Erlotinib and gefitinib for treating non-small cell lung cancer that has progressed following prior chemotherapy (review of NICE technology appraisals 162 and 175): a systematic review and economic evaluation

Diagnosis and staging

Performance status

The measure of PS indicates the degree of a patient’s general well-being. The PS rating may be used when determining fitness for treatment, need for dose adjustment and a patient’s supportive care needs. The three main PS scales are the World Health Organization (WHO) PS scale,8 the Eastern Cooperative Oncology Group (ECOG) PS scale9 and the Karnofsky PS (KPS) scale. 10 The WHO PS scale is most commonly used in UK clinical practice and is described in Table 4. A WHO rating of 0 indicates that a patient is completely able to look after him/herself and a rating of 4 indicates that a patient requires substantial support.

Epidemiology

Relevant national guidelines, including National Service Frameworks

Clinical guidelines published by NICE (NICE CG1216) provide recommendations for good practice in the diagnosis and treatment of lung cancer in England and Wales. In addition, NICE has published a quality standard (NICE QS1715) that defines best practice for the care of people with lung cancer. The QS1715 states that people with stage IIIB or IV NSCLC and eligible PS scores should be offered systemic therapy (first and second line) in accordance with NICE guidance that is tailored to the pathological subtype of the tumour and individual predictive factors. 16

There are a number of NICE guidance documents that are relevant to this appraisal. These are described in Table 5.

Variation in services and/or uncertainty about best practice

Erlotinib

Erlotinib is available as film-coated tablets in 25 mg, 100 mg or 150 mg. The recommended daily dose of erlotinib is 150 mg taken at least 1 hour after food. No guidance as to duration of treatment is given. Erlotinib is licensed in the UK for the treatment of NSCLC and metastatic pancreatic cancer. The latter indication is not relevant to this review.

In the setting of NSCLC, erlotinib is licensed for use with three patient populations. In the first-line setting, erlotinib is licensed for the treatment of patients with locally advanced or metastatic NSCLC with EGFR activating mutations. The SPC25 stipulates that, prior to initiation of erlotinib therapy, people with chemotherapy-naive NSCLC should undergo EGFR mutation testing using a well-validated and robust methodology.

In the post-first-line maintenance setting, erlotinib is licensed as monotherapy for people with locally advanced or metastatic NSCLC whose disease is stable following four cycles of standard platinum-based first-line chemotherapy.

In the second-line setting, erlotinib is licensed for patients with locally advanced or metastatic NSCLC following failure of at least one prior chemotherapy.

Gefitinib

Gefitinib is available as a 250-mg film-coated tablet. The recommended dose of gefitinib is one 250-mg tablet daily. No guidance as to duration of treatment is given. It is licensed in the UK for the treatment of adult patients with locally advanced or metastatic NSCLC with EGFR-activating mutations. The licence places no restriction on where in the treatment pathway gefitinib is used. As was noted for erlotinib, the SPC26 for gefitinib stipulates that a well-validated and robust methodology is used to determine EGFR mutation status before therapy.

The Special warnings and precautions for use section of the SPC26 notes that an increased incidence of interstitial lung disease has been observed in epidemiological studies of gefitinib. Periodic liver function testing is also recommended for patients treated with gefitinib. The AG is aware that in 2003 the Food and Drug Administration (FDA) in the USA approved the use of gefitinib as a second-line treatment for patients who are refractory to platinum-based chemotherapy or docetaxel. The approval was made under the FDA’s accelerated approval regulations that allow the conditional approval of medicines based on surrogate outcomes, in this case tumour response rate (RR). The manufacturer was then required to provide the FDA with data on survival outcomes. The manufacturer has been unable to provide any data that show a positive benefit of gefitinib for survival and, consequently, the FDA (with the agreement of AstraZeneca) removed the licence for gefitinib use in the USA. 27

Current usage in the NHS

The manufacturer of erlotinib [Roche (UK) Ltd] states in its evidence submission to NICE that 70% of patients who receive second-line treatment receive erlotinib. 28

The manufacturer of gefitinib (AstraZeneca) presented in its evidence submission to NICE the number of patients receiving first-line treatment with gefitinib only. These patients are not relevant to this appraisal.

The pack costs of erlotinib and gefitinib are shown in Table 6. The costs of erlotinib to the NHS are subject to a further (confidential) discount under the patient access scheme.

Search strategies

In addition to searching the two manufacturers’ submissions for relevant references, the following databases were searched for studies of erlotinib and gefitinib:

• EMBASE (via OvidSP), from 1974 to 26 April 2013

• MEDLINE (via OvidSP), from 1946 to 26 April 2013

• The Cochrane Library, from inception to 28 April 2013

• PubMed, from January 2010 to 28 April 2013.

The results were entered into an EndNote X5 (Thomas Reuters, CA, USA) library and the references were de-duplicated. Full details of the search strategies are presented in Appendix 1.

The reference lists of included trials were searched for relevant trials. Information on trials in progress was sought from oncology conference databases (American Society for Clinical Oncology and the European Society for Medical Oncology), and the ClinicalTrials.gov website was searched for ongoing trials. In addition, advice was sought from the two clinical advisors to the review and the three clinical peer reviewers.

Inclusion and exclusion criteria

Two reviewers, JG and JH, independently screened all titles and abstracts identified via searching and obtained full-paper manuscripts that were considered relevant by either reviewer (stage 1). The relevance of each study was assessed (JG and JH) according to the criteria set out in this section (stage 2). Studies that did not meet the criteria were excluded and their bibliographic details were listed alongside reasons for their exclusion. Any discrepancies were resolved by consensus and, where necessary, a third reviewer was consulted.

Data extraction strategy

Data relating to both study design and quality were extracted by two reviewers (JG and KD) into a Microsoft Excel^® spreadsheet (Microsoft Corporation, Redmond, WA, USA). Two reviewers cross-checked each other’s data extraction and where multiple publications of the same study were identified, data were extracted and it was reported as a single study.

Quality assessment strategy

The quality of clinical effectiveness studies was assessed independently by two reviewers (JG and KD) in accordance with the Centre for Reviews and Dissemination at York University’s suggested criteria. 30 All relevant information is tabulated and summarised within the text of the report. Full details and results of the quality assessment strategy for clinical effectiveness studies are reported in Appendix 2.

Methods of data synthesis

The results of the clinical data extraction and clinical study quality assessment are summarised in structured tables and as a narrative description. For patients who have progressed following prior treatment, the decision problem of interest to this review is made up of the following comparisons: the effectiveness of erlotinib and gefitinib in a population of patients with EGFR M+ tumours, the effectiveness of erlotinib and gefitinib in a population of patients with EGFR M– tumours and the effectiveness of erlotinib and gefitinib in an EGFR unknown population (i.e. whose EGFR mutation status is unknown at the time of randomisation).

Quantity and quality of research available

A total of 1563 titles and abstracts were screened for inclusion in the review of clinical effectiveness and cost-effectiveness evidence. Overall, 12 relevant RCTs were identified. The process of study selection is shown in Figure 1.

FIGURE 1.

Study selection process.

Clinical effectiveness (randomised controlled trials)

A total of 12 RCTs (one of which was discussed in NICE TA162,21 namely BR.2131) were reported in 25 publications and met the criteria for inclusion into the review. The reference cited in the text refers to the primary report, and subsequent publications describing outcomes of the trials are listed by trial in Appendix 3. The AG did not find any relevant publications that were not identified by the manufacturers.

The identified trials are summarised in Table 8. A full list of publications that were excluded from the review following the application of the inclusion criteria is presented in Appendix 4. The AG also identified and assessed the quality of existing systematic reviews in order to cross-check for the identification of additional studies as well as to gain an understanding of the issues related to the combining of data in this complex clinical area. A summary and critique of relevant systematic reviews is presented in Appendix 5.

As EGFR mutation status is a key factor in this review, it is noted in Table 8 whether or not a patient’s EGFR mutation status was determined before randomisation and used as the basis for inclusion in the trial. For those trials that did not select patients based on EGFR mutation status, the final column of the table indicates whether or not any retrospective analyses of the data were conducted. It should be noted that, where retrospective EGFR subgroup analyses are available, the data are limited.

Two of the included trials, Bhatnagar et al. 33 and the Docetaxel and Erlotinib Lung Cancer Trial (DELTA),40 were only reported as conference abstracts and, therefore, limited information is available to describe these studies. The final results of the TArceva Italian Lung Optimization tRial (TAILOR)41 were published after our searches were completed; however, we have included these results in the review.

Assessment of effectiveness

The AG’s assessment of effectiveness is based on the following patient groups:

1. previously treated adult patients with locally advanced or metastatic NSCLC and who exhibit EGFR-activating mutations (referred to as EGFR M+ population)

2. previously treated adult patients with locally advanced or metastatic NSCLC and who do not exhibit EGFR-activating mutations (referred to as EGFR M– population)

3. previously treated adult patients with locally advanced or metastatic NSCLC and for whom EGFR mutation status is unknown or indeterminate (referred to as EGFR unknown population).

Epidermal growth factor mutation-positive population

• No trials were identified that were conducted in a population of solely EGFR M+ patients. Limited EGFR mutation status data were retrospectively derived from relatively small subgroup analyses of RCTs that included patients of unknown EGFR mutation status at the time of randomisation.

• Four studies reported OS outcomes,31,34,39,42 none of which was statistically significantly different for any of the comparisons described.

• Five studies reported PFS,31,32,34,39,42 but only one trial36 found a statistically significant improvement for any comparison considered, and the results favoured gefitinib over docetaxel.

Epidermal growth factor mutation-negative population

• Key data were derived from results of TAILOR41 and DELTA40 trials.

• EGFR mutation status data were retrospectively derived from subgroup analyses in BR.21,31,43 Kim et al. ,32 TITAN,42 INTEREST,34,45 and ISEL. 39,44

• OS outcome: no statistically significant differences were noted for OS for either erlotinib or gefitinib compared with any treatment.

• PFS outcome: TAILOR41 and DELTA40 reported a statistically significant benefit of docetaxel compared with erlotinib. No statistically significant PFS benefit was reported from subgroup data.

• RR: patients in the docetaxel arm of TAILOR41 had statistically significantly higher RRs than patients in the erlotinib arm.

Epidermal growth factor mutation unknown: overall population

• Data were available from 11 trials31–41 carried out in populations in which EGFR mutation status was not a factor in the recruitment process (or in which overall trial results were presented).

• OS outcome: the only statistically significant OS benefit for any treatment was reported in BR.2131 (erlotinib vs. placebo). However, this finding was based on an adjusted rather than an unadjusted analysis of the data.

• PFS outcome:

  • Gefitinib versus docetaxel – only one of the four trials (ISTANA35) reported a statistically significant benefit of gefitinib.

  • Gefitinib versus BSC – gefitinib was reported to have a statistically significant benefit. 39

  • Erlotinib versus placebo (BR.2131) – a statistically significant PFS benefit of erlotinib was reported (in an adjusted analysis).

• RR: of the trials reporting RRs,31,32,34–39,41 two noted significant differences in favour of gefitinib when compared with docetaxel38 and BSC. 39

Meta-analysis and network meta-analysis

For clinical and methodological reasons, no meta-analysis or network meta-analysis was conducted by the AG.

Quality of life

Where reported, the QoL data were derived from the EGFR unknown patients (overall population, i.e. the data are not specific to the EGFR mutation status of patients). All of the 12 trials included in this review measured QoL. However, the QoL outcomes from TAILOR41 and DELTA40 are not yet available.

Adverse events

Adverse events were reported for the overall population, that is the data are not specific to the EGFR mutation status of patients, with the exception of TAILOR. 41 Details of the AEs reported in Bhatnagar et al. ,33 Li et al. 36 and DELTA40 were limited. The AG considers that the AEs reported, despite inconsistencies across trials, appear to be consistent with the information available for erlotinib, gefitinib and docetaxel in the SPCs. 24

Erlotinib

Clinical evidence supporting the previously published NICE guidance TA16221 (erlotinib for the treatment of NSCLC) issued in 2008 was based on the results of a single RCT, the BR.2131 trial, that compared erlotinib with placebo. At the time of the appraisal of erlotinib in NICE TA162,21 no direct evidence comparing erlotinib with docetaxel was available and, in the evidence submission to NICE, the manufacturer of erlotinib presented an indirect treatment comparison in which docetaxel was compared with BSC and pemetrexed. The Appraisal Committee (AC) did not consider the indirect treatment comparison to be robust and concluded that it was difficult to reach a decision as to the effectiveness of erlotinib compared with docetaxel. NICE guidance (TA16221) states that erlotinib is recommended, within its licensed indication, as an alternative to docetaxel as a second-line treatment option for patients with NSCLC only on the basis that it is provided by the manufacturer at an overall treatment cost (including administration, AEs and monitoring costs) equal to that of docetaxel. The patient access scheme was then superseded to a simple discount patient access scheme following the publication of NICE TA227. 47 The price of erlotinib relevant to the NHS now is that of the list price minus the simple discount, as noted in the latest version of NICE TA162. 21

Since the publication of NICE TA162,21 three developments are worthy of note. First, the results of one RCT comparing erlotinib with chemotherapy (TITAN42) in a population of patients with unknown EGFR status have been published. The chemotherapy comparator was docetaxel or pemetrexed according to the treating physician’s choice. Pemetrexed is licensed as a second-line treatment but is not recommended by NICE and therefore was not listed as a comparator in the decision problem for this appraisal. No statistically significant differences between erlotinib and chemotherapy were reported. The authors of the published paper42 note that the choice of either docetaxel or pemetrexed was at the treating physician’s discretion and treatments were, therefore, not randomised. In addition, pemetrexed and docetaxel were not always available in all centres. For these reasons, the trial investigators published only outcomes for chemotherapy (i.e. aggregated), as the results of efficacy of erlotinib versus docetaxel and erlotinib versus pemetrexed were considered unreliable.

Second, the patent for docetaxel has expired. Docetaxel is now available generically at a considerably reduced price (less than 10% of its previous list price). 48 To date, NICE has not issued any statement suggesting that this lower price of docetaxel necessitates any change to the recommendations set out in NICE TA162. 21

Third, clinical practice has also changed since the publication of NICE TA162,21 with the identification of EGFR mutation status as a prognostic factor. Erlotinib is an EGFR-TKI and is licensed as a first-line treatment for patients with EGFR M+ tumours and as a second-line treatment for locally advanced or metastatic NSCLC regardless of EGFR mutation status. As noted previously, the majority of patients in clinical practice in England and Wales have their tumours histologically tested at diagnosis and prior to first-line treatment. Patients who are likely to have EGFR M+ tumours are also tested for activating mutations. Patients who test positive for EGFR-activating mutations are treated at first line with a TKI (either erlotinib or gefitinib), while those who are EGFR M– are treated with third-generation platinum-doublet chemotherapy or monotherapy. On progression, EGFR M+ patients are not re-treated with an EGFR-TKI and therefore receive docetaxel in line with current NICE guidance. 21 The AG is aware that some patients in the NHS are given platinum-doublet chemotherapy after first-line EGFR-TKI; however, this treatment pathway is not standard UK clinical practice. Patients who are EGFR M– are offered erlotinib or docetaxel. In summary, increased significance of EGFR mutation status in lung cancer treatment raises questions about how to treat both EGFR M+ and EGFR M– patients.

Two recent trials (TAILOR41 and DELTA40) were both designed to compare the effectiveness of erlotinib versus docetaxel in EGFR M– patients. The results of TAILOR41 are reported in a published paper, while the results of DELTA40 are presently available only as a conference abstract from the American Society for Clinical Oncology in 2013. Since TAILOR41 provides key data on the effectiveness of erlotinib compared with docetaxel in the EGFR M– population, further consideration of the trial and its relevance to clinical practice in England and Wales is warranted here.

TAILOR41 was conducted in 52 hospitals in Italy and randomised patients to receive erlotinib (n = 112) or docetaxel (n = 110). While OS was not statistically significantly different between the two arms, there was a statistically significant benefit of docetaxel over erlotinib for PFS. The QoL data are not yet available.

TAILOR41 has attracted a number of criticisms. First, the primary objective of the trial was changed at the first planned interim analysis. According to the published paper,41 the trial was initially designed to assess the effects of docetaxel and erlotinib according to the biomarkers of EGFR amplification and protein expression, and KRAS mutations. When, after masked efficacy analysis, these biomarkers were found to have no effect, the independent monitoring and safety committee recommended that the primary objective of the trial be changed to a comparison of efficacy between erlotinib and docetaxel with a primary end point of OS.

Second, TAILOR41 employed two regimens of docetaxel administration, either 75 mg/m² every 3 weeks or weekly infusions of 35 mg/m². The AG notes that this latter regimen would not be used in clinical practice in England and Wales.

Third, the fitness of the patients in TAILOR41 is an important consideration. The patient population consisted of a majority of patients with an ECOG PS score of 0 or 1 and only 7% with a PS score of 2. This is unlikely to reflect patients in the NHS, in which a higher proportion of PS 2 patients would be treated in routine clinical practice. The AG is aware that PS is a prognostic factor in NSCLC and poorer PS is linked to poorer outcomes. However, the AG notes that the patient population in TAILOR41 may reflect future populations of patients seen in clinical practice in England and Wales as treatment for NSCLC continues to evolve. In modern clinical practice, patients are diagnosed earlier and treated more aggressively than in the past, which means that future patients may be fitter at second line than those currently receiving second-line treatments in England and Wales.

Fourth, there are differences in other important prognostic factors between the treatment arms of TAILOR. 41 There are differences in patient characteristics (docetaxel vs. erlotinib): never-smokers (27% vs. 17%), squamous cell (21% vs. 28%) and adenocarcinoma (75.55% vs. 63.00%). All of these differences have been identified as possible modifiers of trial outcome in favour of docetaxel. 48

In its submission to NICE, the manufacturer of erlotinib has questioned the low rates of haematological toxicity in the docetaxel arm of TAILOR41 [febrile neutropenia (FN) grade 3 or 4 = 4%, neutropenia grade 3 or 4 = 21%] in comparison with the INTEREST34 trial (FN grade 3 or 4 = 10%, neutropenia grade 3 or 4 = 58%) and the JMEI49 trial (FN grade 3 or 4 = 13%, neutropenia grade 3 or 4 = 40%). The manufacturer questions whether or not these low rates are related to the fitter patient population or the use of weekly treatment schedules. The AG considers that there may be another explanation, the increased clinical awareness of docetaxel-related AEs. Docetaxel has been used in the NHS for many years and it is likely that these related AEs are currently better managed and/or more frequently avoided than in the past.

In summary, it is open to debate how far TAILOR41 reflects clinical practice in England and Wales and, therefore, whether or not the trial results are likely to be mirrored in a UK clinical population. TAILOR41 is a large, high-quality RCT in a population of patients who do not have activating EGFR mutations. The trial is very relevant to patients in the UK, as it compares two lung cancer treatments that are currently recommended by NICE for the post-progression treatment of patients with NSCLC.

The specific details of DELTA40 are as yet unavailable and so it is not possible to assess how far the Japan-based trial reflects clinical practice in England and Wales.

Gefitinib

In 2009, NICE was unable to recommend the use of gefitinib in the NHS for the second-line treatment of locally advanced or metastatic NSCLC because no evidence submission was received from the manufacturer or sponsor of the technology. 22

The marketing authorisation for gefitinib granted by the European Medicines Agency50 was based on the results of the first-line IRESSA Pan-Asia Study51 trial and second-line INTEREST34 trial. Supporting trials included ISEL,39 SIGN,37 V-15-3238 and ISTANA. 35 The European Medicines Agency’s European Public Assessment Report52 reports that concerns were raised by the scientific advisory group about the data submitted by AstraZeneca (London, UK) in support of the licensing application for gefitinib. In particular, the advisory group noted a large number of missing data with respect to EGFR mutation status and considered that this aspect should have been controlled for by the design and conduct of the clinical studies. In this respect, the clinical studies presented were considered by the European Medicines Agency52 to be inadequate. Three new trials of gefitinib have been published since 2009, which was when the European Medicines Agency52 considered the application. The three trials were conducted in small populations of patients, Kim et al. 33 (vs. erlotinib), Li et al. 36 (vs. docetaxel) and Bhatnagar et al. 33 (vs. docetaxel), and the new data they provide are not sufficiently robust to permit recommendation of a change in clinical practice.

The AG notes, as does the manufacturer of gefitinib, that in clinical practice in England and Wales patients with EGFR M+ NSCLC should be diagnosed and treated appropriately (with a TKI) at first line. As noted above, patients who go on to second-line treatment will not be re-treated with the same therapy. It is likely, therefore, that the number of patients treated with gefitinib after progression will be limited to a very small minority who were not treated with a TKI at first line, perhaps as a result of lack of diagnostic facilities.

Meta-analysis and network meta-analysis

In view of the paucity of relevant data, the AG was unable to conduct either a meta-analysis or network meta-analysis in respect of the efficacy of treatments for patients with known EGFR M+, EGFR M– or EGFR unknown NSCLC.

The majority of the clinical evidence lies with the trials that included patients with NSCLC who were of unknown mutation status. Unfortunately, a number of issues precluded any comparison of the available data for patients with NSCLC of unknown mutation status, the issues were both clinical (differences in patient populations) and methodological (adjusted vs. unadjusted outcome data, Cox proportional hazards violations). However, even if the comparison could have been carried out, given the increased significance of EGFR mutation testing, its relevance to the current decision problem and to modern clinical practice is questionable.

From the 12 included RCTs, the most reliable evidence is from a study of the EGFR M– population. For this group of patients, the results of TAILOR41 demonstrate that there is a statistically significant benefit of docetaxel over erlotinib for PFS; however, there is no statistically significant OS benefit demonstrated in this trial.

Methods of cost-effectiveness review

Full details of the main search strategy conducted by the AG and the proposed methods for selecting clinical and economic evidence are presented in detail in Chapter 3, Methods for reviewing effectiveness. The AG did not use specific economics-related search terms in the main strategy, as all of the potential references were scanned for references containing economic evidence. For the selection of cost-effectiveness evidence, AB and SB independently screened all economics-related titles/abstracts identified via searching and obtained full-paper manuscripts of all relevant references. The relevance of each study was then assessed (by AB and SB) according to the specific inclusion and exclusion criteria shown in Table 23. Data were extracted (AB and SB) and summarised in structured tables and as a narrative description.

In the NHS in England and Wales (and in countries elsewhere in the world), docetaxel is commonly used to treat patients with NSCLC who have progressed after chemotherapy and is, therefore, described as a relevant comparator to erlotinib and gefitinib in published economic evaluations (EEs). Recently, the price of docetaxel has fallen29 substantially as a result of the expiry of the manufacturer’s patent. The AG discussed whether or not to exclude papers that presented data using the higher docetaxel price. The AG decided to include these papers but to highlight in the discussion section that the results of EEs that only include docetaxel at its higher price are of limited relevance to this appraisal.

Until recently, patients who required post-progression treatment for NSCLC were treated as a homogeneous group. However, clinical practice is now changing and there is growing awareness that a patient’s EGFR mutation status can affect treatment outcomes. With this in mind, the AG discussed excluding papers that did not consider how EGFR mutation status can affect patient outcomes and the treatment options available. However, on reflection, the AG decided not to exclude these papers but to highlight in the discussion that the results of EEs that only include patients with EGFR unknown status should be treated with caution.

Quantity of included evidence

From the main search, the AG identified 44 potentially relevant economic papers for inclusion in the review of economic evidence. Of these, 16 papers were considered for inclusion after stage 1 screening. Of these 16 papers, 10 were then excluded from the review and six were included in the review at stage 2. The reasons for excluding 10 papers are listed in Table 24.

From the systematic review by Bongers et al. ,54 a further four papers were identified for inclusion in the AG’s review. This finding alerted the AG to the fact that the main search had not picked up all of the relevant published economic studies available. The AG then carried out further searching using a combination of the following broad search terms to identify papers in MEDLINE and The Cochrane Library: erlotinib, gefitinib, lung cancer and cost. This additional generic search identified one more relevant paper by Vergnenegre et al. 63

In summary, the AG considered 11 papers to be eligible for inclusion in the review and these are listed in Table 25.

Quality of included evidence

The AG made the decision not to quality assess the papers included in the review of cost-effectiveness evidence. This decision was made because none of the 11 studies is directly relevant to UK health-care decision-making as they do not use the off-patent price of docetaxel. Additionally, none of the studies consider the confirmed EGFR mutation status of the patient when assessing post-progression treatments.

Cost-effectiveness review: results

Relevant data were extracted from the 11 eligible papers (Table 26). These papers were published between 2002 and 2013; seven papers63,65,66,68–71 were published from 2010 onwards. All of the papers described full EEs using cost minimisation analysis (n = 164), cost-effectiveness analysis (n = 665,67,69,70,72) and/or cost–utility analysis (n = 663–65,68,71,73) techniques. All but one study70 used cost per QALY gained or cost per LY gained as the measure of cost-effectiveness. The results of six studies64,65,67,68,71,73 were derived from use of an economic model: one study63 conducted an economic analysis alongside a RCT and the remaining four studies66,69,70,72 conducted retrospective reviews of costs and/or benefits. Four studies66,69,70,72 were carried out from a Canadian NHS perspective, two67,71 from that of the UK NHS perspective, one73 from the US perspective, three63–65 from a European perspective and one68 from a Thai payer perspective. None of the studies had a time horizon of longer than 3 years. The authors of two studies69,70 had not received any financial support from the pharmaceutical industry.

The 19 comparisons described in the 11 economic studies included one or more of the following interventions: erlotinib, docetaxel, pemetrexed and BSC. The most common comparison was erlotinib versus docetaxel (n = 564,68,69,71,73). Other comparisons were erlotinib versus BSC (n = 364,66,70), pemetrexed versus docetaxel (n = 463,65,68,73), docetaxel versus BSC (n = 363,67,72), erlotinib versus pemetrexed (n = 264,73), pemetrexed versus BSC (n = 163) and gefitinib versus docetaxel (n = 168). The populations described in the EEs appeared to have similar patient characteristics, namely previously treated stage III–IV patients with advanced NSCLC. The clinical data used in the EEs were derived mainly from relevant published RCT data: TAX31774 (docetaxel vs. BSC), JMEI49 (pemetrexed vs. docetaxel), BR.2131 (erlotinib vs. placebo) and INTEREST34 (gefitinib vs. docetaxel). The source of the clinical data described in two studies was patient medical records. The paper by Nafees et al. 75 provided the source of the QALY values in two papers. 63,73

The outcome data (e.g. QALY values and LYs gained) used in the evaluations were variable as a result of the assumptions employed (Table 27). To illustrate, the average total QALY value accrued over the time horizon of the models associated with each of the drugs used in the studies range was as follows: erlotinib (0.17468 to 0.42075), docetaxel (0.16068 to 0.42073), pemetrexed (0.17168 to 0.52065). In addition, the AG notes that Araujo et al. 64 assume that erlotinib, docetaxel and pemetrexed yield equivalent LYs (0.77 years), Thongprasert et al. 68 assume that the gain in LYs is equivalent when comparing docetaxel and pemetrexed (0.97 years) and when comparing erlotinib and gefitinib (0.96 years), and Carlson et al. 73 assume that the gain in LYs for erlotinib, docetaxel and pemetrexed is equivalent (0.77 years).

Cost data were mainly derived from relevant national sources of published cost information (Table 28), for example Spanish reference database (BOT database of pharmaceutical prices),65 Portuguese ministerial dispatch report,64 Ontario Case Costing Acute Inpatient Database69 and British National Formulary (BNF). 67 Costs were typically categorised as drug, drug administration and/or monitoring and treatment of AEs. The publication year differed by no more than 3 years from the base-cost year used in the studies.

The costs estimated and employed in the EEs differ because of the assumptions made by the authors. For example, total costs per patient for erlotinib range from CA$16,487 to CA$35,708. 69 In Vergnenegre et al. ,63 the costs of BSC are assumed to equal zero while in Leighl et al. 72 the average cost of care in the BSC group was CA$6935.04. Costs and benefits were discounted at a 3%, 3.5% or a 5% discount rate, although some studies69,71,72 did not use discounting despite estimating costs and benefits over a time period greater than 12 months.

Despite variations in the methods employed and reporting of results across the studies, five of the six studies that assessed erlotinib compared with chemotherapy or BSC favoured erlotinib;64,66,70,71,72 the authors of the remaining study69 concluded that erlotinib and docetaxel were equal in terms of costs and benefits. Two studies67,72 comparing docetaxel versus BSC concluded that docetaxel was cost-effective. In another study68 gefitinib was preferred to docetaxel, and, of the two studies comparing pemetrexed versus docetaxel, one study favoured docetaxel63 and the other favoured pemetrexed. 65

Cost-effectiveness review: discussion of study methods and results

It is clear from the methods and results reported in the published cost-effectiveness literature that the conclusions drawn are very dependent on the assumptions made by the investigators and the data sources employed in the EEs (Table 29). These differ from evaluation to evaluation. Each EE must therefore be judged on its own merits and any attempt to make summary statements about different comparisons in terms of cost-effectiveness is meaningless.

Of the 19 comparisons considered in the 11 published studies, 13 included docetaxel as a comparator. The AG notes that the patent on docetaxel has expired and docetaxel is now available in its generic form at a cost that is less than 10% of its previous list price. 29 The AG therefore considers that the incremental cost-effectiveness ratios (ICERs) estimated in these 13 comparisons are now of limited value to decision-makers in the UK NHS. Of the six remaining comparisons, three included pemetrexed as a comparator [pemetrexed vs. BSC (n = 163) and pemetrexed vs. erlotinib (n = 264,73)]. Again, the AG considers that the results of these studies cannot be used directly to inform decision-making in the UK as pemetrexed is not recommended by NICE for the second-line treatment of patients with NSCLC in the UK NHS. The remaining three studies64,66,69 focused on the comparison of erlotinib with BSC. However, as none of the studies report ICERs for an EGFR M+ or EGFR M– patient population, the AG considers that the estimated ICERs are useful only when making treatment decisions for patients whose EGFR status is unknown, as the EGFR mutation status of this patient group can influence treatment choices. In addition, the AG is of the opinion that, although BSC is a valid comparator for a small population of patients with NSCLC, docetaxel is a more appropriate comparison for patients in the UK NHS.

The AG concludes that the results of the systematic review are of limited value to decision-makers in the UK NHS. This is a result of (1) relatively recent changes in the price of docetaxel and (2) the increased significance of EGFR mutation testing for patients with NSCLC. The AG does not summarise or draw conclusions from any other manufacturer’s submission used in previous NICE appraisals of erlotinib and/or gefitinib as these submissions were written at a time when it was not possible to take into account these aforementioned changes. The AG anticipates that future EEs in this complex clinical area will make use of the most up-to-date clinical effectiveness and cost data available.

Review of Roche (UK) Ltd economic model: erlotinib versus best supportive care

The ERG assessed the economic model submitted by Roche using NICE’s reference case checklist (Table 30).

Overview of submitted manufacturer’s submission

The manufacturer developed a de novo economic model using data from the BR.2131 trial. In the base-case analysis, the manufacturer compares erlotinib versus BSC using ITT data from the BR.2131 trial. In a separate subgroup analysis, the manufacturer compares erlotinib versus BSC in an EGFR M– patient population only, this patient group was identified retrospectively. 43

The developed model is a partitioned survival model with three health states (a structure that has been used in many previous NICE oncology technology appraisals, including TA162,21 TA22747 and TA29576). The model projects PFS and OS independently with the proportion of patients in the progressed health state over time being the proportion of patients alive but not in the PFS health state.

The model structure is shown in Figure 2. All patients enter the model in the PFS health state and in each month can either progress to a worse health state [i.e. from PFS to progressed disease (PD) or from PD to death] or remain in the same health state. The model has been developed in Microsoft Excel and has a 1-week cycle length.

FIGURE 2.

Schema of manufacturer’s model. PD, progressed disease.

Population

The population was assumed to be the same as that recruited to the BR.2131 trial, that is patients 18 years of age or older with an ECOG PS score of between 0 and 3 and who had documented pathological evidence of NSCLC. Patients in this trial had to have received one or two regimens of combination chemotherapy and not be eligible for further chemotherapy. The only baseline population characteristic used in the model was age (61.4 years in both arms).

Interventions and comparators

The manufacturer believes that, following the availability of generic docetaxel at less than 10% of the previous list price, it is not possible to demonstrate that erlotinib is cost-effective when compared with docetaxel. The manufacturer has, therefore, only presented an analysis comparing erlotinib (maximum of one 150-mg tablet per day until disease progression) with BSC. In addition, the AG notes that the manufacturer did not compare the cost-effectiveness of erlotinib with gefitinib.

Perspective, time horizon and discounting

The EE is undertaken from the perspective of the NHS and Personal Social Services Research Unit (PSSRU). Outcomes are expressed in terms of LYs gained and QALYs gained. The time horizon is set at 6 years and, in line with the NICE Guide to the Methods Technology Appraisal,77 both costs and benefits are discounted at 3.5%.

Treatment effectiveness and extrapolation

Data from BR.2132 were used to estimate PFS and OS.

Health-related quality of life

The manufacturer extracted utility values from the published appraisal of crizotinib for the treatment of previously treated NSCLC associated with a lymphoma kinase fusion gene (NICE TA29678). The manufacturer selected and applied the pooled chemotherapy (pemetrexed or docetaxel) values to both the erlotinib and BSC arms of the model. The manufacturer considers this to be a conservative assumption, as QoL data from BR.2131 showed that erlotinib improved QoL as regards time to deterioration of key symptoms of cough, dyspnoea and pain compared with BSC.

The manufacturer notes that the patient population in PROFILE 100779 (described in TA29678) is anaplastic lymphoma kinase positive and that the utility values from this population are relatively high for patients with NSCLC. Furthermore, the patient group in PROFILE 100779 was younger and less fit than those patients enrolled in the BR.2131 trial.

The source of the utility values used in the model is presented in Table 31.

Resources and costs

Cost-effectiveness results

The base-case incremental results generated by the manufacturer’s model are presented in Table 35. The ICER for the comparison of erlotinib with BSC in patients with NSCLC whose EGFR mutation status is unknown and who have progressed after prior chemotherapy treatment, is £51,036 per QALY gained and £35,593 per LY gained. Disaggregated costs for the target population are presented in Table 36.

Sensitivity analyses

The manufacturer carried out a large number of one-way sensitivity analyses. A tornado diagram is included in the manufacturer’s submission (figure 27, page 67). The one-way sensitivity analysis results for the five changes that have the largest impact on cost-effectiveness are displayed in Table 37.

Probabilistic sensitivity analysis (PSA) was undertaken (5000 iterations of the model) by the manufacturer. A scatterplot (incremental cost vs. QALY) and a cost-effectiveness acceptability curve are included in the manufacturer’s submission (p. 70) and reproduced in Figures 3 and 4.

FIGURE 3.

Probabilistic sensitivity analysis scatterplot erlotinib vs. BSC (diagonal line = £30,000 per QALY gained).

FIGURE 4.

Cost-effectiveness acceptability curve.

The results of the PSA are displayed in Table 38. The PSA ICER is estimated to be £50,825 per QALY gained, which is only £211 less than the base-case deterministic ICER of £51,036 per QALY gained.

The PSA results show that there is a 0% probability that erlotinib is cost-effective at a willingness-to-pay threshold of £30,000 per QALY gained. However, at a threshold of £60,000 per QALY gained there is a 40% probability that erlotinib is cost-effective, and at a threshold of £65,000 per QALY gained erlotinib is cost-effective in approximately 76% of all scenarios.

Subgroup analysis

The manufacturer undertook a separate subgroup analysis for the EGFR M– population of the BR.2131 trial using data from the publication by Zhu et al. 43 The ICER for this group was £58,579 per QALY gained, a value which is approximately 14% higher than the base-case ICER. The QALY gain comes entirely from the PFS health state. The manufacturer advises that the results from this analysis, which are displayed in Table 39, should be interpreted with caution because of the limitations of the available data.

Critique of the submitted model

The AG notes that, as well as not analysing the cost-effectiveness of erlotinib compared with docetaxel, the manufacturer did not carry out an analysis of the cost-effectiveness of erlotinib compared with gefitinib. This critique therefore focuses on the manufacturer’s analysis of the cost-effectiveness of erlotinib compared with BSC that is presented in the manufacturer’s submission. A detailed examination of model formulae and calculations has not been carried out.

The economic model submitted by the manufacturer was of a structure used in many previous oncology technology appraisals. The presented evaluation was based on data from one RCT (BR.2131). This trial recruited a population of patients with NSCLC and of unknown EGFR status; however, treatment pathways have evolved and currently patients who have EGFR M+ disease would not generally be given a EGFR-TKI as a second-line treatment, as they would already have received a TKI as a first-line therapy.

The manufacturer carried out a wide range of sensitivity analyses. The biggest impact on the size of the cost per QALY ICER (an increase of £10,281) resulted when utility values from Nafees et al. 75 replaced values from PROFILE 100779 in the manufacturer’s base-case analysis.

The AG has several concerns about the use of PROFILE 100779 values in the base-case analysis, namely:

• These values have not been published, peer reviewed or validated.

• There is no information on the coverage of patients within the trial completing the survey (i.e. at which time point and at which stage of treatment) so no assessment can be made of the potential for bias in any overall averages obtained.

• The crude averages incorporate the effects of treatment-related AEs, which relate to another treatment given to younger but less fit patients with a different type of NSCLC.

In the manufacturer’s economic model, the social tariff algorithm used to calculate European Quality of Life-5 Dimensions (EQ-5D) scores is unknown. As the predominant data source in the PROFILE 100779 trial is the USA, it would not be surprising if the US tariff, which gives consistently higher scores than the UK tariff, had been used.

Figure 5 shows the relationship between health state scores using UK and US tariffs. When this conversion is applied to the PROFILE 100779 utility scores, the PFS average (US tariff) changes. The Nafees et al. 75 model gives 0.653 for stable disease PFS and 0.673 for responder PFS. Similarly, the PD average US tariff utility changes and compares closely with the Nafees et al. 75 PD utility.

FIGURE 5.

Relationship between health-state scores using UK and US tariffs.

One further point which, in this case, is likely to have only a minor impact on the size of the cost per QALY ICER relates to the cost of a hospital pharmacist’s time, which is used to estimate erlotinib administration costs. A value of £91 per hour (PSSRU 201181) has been used in the model but the most up-to-date value is £67 (PSSRU 201284).

In view of these issues, and to allow all therapy options to be compared using a consistent framework, the AG has developed a de novo cost-effectiveness model.

Methods

Model design

The decision model (Figure 6) is conceptually straightforward, involving two health states prior to death (progression free after second-line chemotherapy, post progression). Therapy is treated as an extended event, given over several cycles (usually of 3 weeks’ duration). However, orally administered treatments (erlotinib and gefitinib) are given continuously until the disease progresses, and treatment is assumed to be coterminous with the duration of the PFS state.

FIGURE 6.

Conceptual model of second-/third-line decision model, indicating health states (rectangles), events/procedures (ovals) and transitions (arrows).

Disease progression after second-line therapy is treated as an event, resulting in one of two transitions, either to a period of post-progression survival (PPS), which eventually results in death, or to immediate death. Further lines of therapy are possible but are not modelled explicitly, as the proportion of patients receiving subsequent active treatments is small in the UK. Instead, additional resources and utility effects are included in the post-progression health state to represent average usage.

The model is implemented as a Microsoft Excel workbook, using macro-programming to perform PSA to assess the relative probabilities of cost-effectiveness between the available second-line treatments.

Ideally, the model should be driven by evidence from clinical trials relating to each of the model’s health states: the duration of PFS when patients receive second-line treatment and the duration of PPS when patients receive only BSC. Unfortunately, the only outcomes routinely reported for clinical trials are PFS and OS. Thus the model can only be populated indirectly by inferring the probable experience of patients in the intermediate states. This leads to potentially serious difficulties and inconsistencies in model implementation. In particular, the normal practice of treating PFS and OS as independent variables is naive, since PFS is a major component of OS. Not recognising this easily leads to situations where deriving an estimate for PPS by subtracting estimated PFS from estimated OS leads to erroneous negative values at some point during the simulation period. The modeller has to exercise great care at every stage of model development, calibration and use so as to guard against producing nonsensical results.

Synthesis of outcome data: progression-free survival and overall survival

Epidermal growth factor mutation-positive population

No clinical trials have been identified which compare second-line treatments in a population of only patients with EGFR-activating mutations.

Epidermal growth factor mutation-negative population

Clinical effectiveness data for this patient group are restricted to TAILOR,41 which compares erlotinib with docetaxel. Published Kaplan–Meier survival curves were digitised by the AG to provide source data for projecting the full cohort experience until death. Both PFS and OS curves exhibited forms inconsistent with the standard parametric functions routinely featured in commercially available statistical software. All such functions assume that a single continuous disease and treatment process is in effect throughout the duration of the trial, resulting in gradual smooth changes in event risk and survival outcomes from randomisation until the outcome event (progression/death for PFS or death for OS). The Kaplan–Meier curves from TAILOR41 show clearly that this assumption is invalid, with quite different behaviour exhibited over different periods of the trial in both patient groups.

The natural history of untreated advanced/metastatic lung cancer is generally straightforward, involving a high but constant risk of disease progression and death within a short time period (usually best represented as a Poisson process, i.e. an exponential survival function). However, when short-term interventions are applied to patients, the normal disease dynamic is distorted, typically into three time periods: an initiation period (prior to treatments achieving full efficacy), an efficacious period (when different treatments may show divergent risk of progression/death) and a loss of efficacy period (when the natural course of progressive disease is reasserted).

Examination by the AG of the cumulative hazard plots for the trial data indicated that a three-phase spline model (with two ‘knot’ points) closely represents the published trial results and outperforms any of the standard parametric functions conventionally employed. In the first phase, event risks are very similar in both trial arms. In the second phase, patients in both trial arms are subject to increased risk of an event (progression or death) and at different levels of risk corresponding to differential treatment efficacy, so that the survival curves diverge. In the final phase, event risks reduce substantially in both arms. In addition, the transitions between phases appear to occur at the same time from randomisation in both treatment arms. The event risk within each phase was found to conform closely to a constant (equivalent to an exponential survival function) in both treatment arms. The main structural difference between statistical models for the two treatments occurs in the final phase. For PFS the event risk remains higher in the erlotinib arm, suggesting that PFS outcomes continue to diverge indefinitely, whereas in the OS comparison the long-term mortality risk stabilises at the same level once all patients have suffered disease progression, thus suggesting that for the remainder of patients’ lifetimes survival prognosis is unrelated to previous treatments.

Figures 7 and 8 demonstrate the correspondence between TAILOR42 data and the AG’s projective models. The calibrated models were only used to project PFS and OS during and beyond the third phase to maximise the use of the unadjusted trial data. In all cases projection was commenced at the same value of the estimated remaining PFS or OS to avoid introducing bias from projecting different proportions of patient experience subject to different degrees of modelling error. For PFS projection began at the point when 30% of patients were estimated to be event-free and for OS projection began at 41%. Details of the AG’s model parameters, estimates and standard errors are provided in Appendix 7.

FIGURE 7.

Three-phase projective spline models fitted to PFS data from TAILOR. 42

FIGURE 8.

Three-phase projective spline models fitted to OS data from TAILOR.

Epidermal growth factor mutation status unknown population

Clinical effectiveness data for this patient group are restricted to the BR.2131 trial. The manufacturer’s model included detailed Kaplan–Meier analysis data, which provided the source data for projecting the full cohort experience until death. Both PFS and OS curves exhibited similar forms to those observed in TAILOR. 41 Therefore, a similar three-phase spline model (with two ‘knot’ points) was employed for analysis of the BR.2131 data. The transitions between phases (‘knot’ points) in the two trial arms occur at different points between the first two phases but at a common time point between phases 2 and 3. The event risk within each phase was found to conform closely to a constant (equivalent to an exponential survival function) in both treatment arms. In both OS and PFS models the long-term event risk (phase 3) exhibits the same hazard rate in both arms of the trial.

In these circumstances a simplified model formulation could be focused on the final long-term period (phase 3), recognising that accurate Kaplan–Meier data are available into the final period and should be applied directly, which limits the need for projection of missing data to a short final period. A single exponential long-term model was calibrated for a single hazard parameter, and separate constant parameters for each treatment arm which together correspond to the separation between the survival curves at the second ‘knot’ point (296 days).

Figures 9 and 10 show the correspondence between the trial data and the late-stage projective models. These calibrated models were only used to project PFS and OS during and beyond the third phase to maximise the use of the unadjusted trial data. In all cases projection was commenced at the same value of the estimated remaining PFS or OS to avoid introducing bias from projecting different proportions of patient experience subject to different degrees of modelling error. For PFS projection began at the point when 5% of patients were estimated to be event-free and for OS at 25%. Details of the model parameters, estimates and standard errors are provided in Appendix 7.

FIGURE 9.

Long-term projective models fitted to PFS data from the BR.21 trial. 31

FIGURE 10.

Long-term projective models fitted to OS data from the BR.21 trial. 31

Results

Epidermal growth factor mutation-negative population

Erlotinib versus docetaxel

Deterministic results from the main EGFR M– model based on data from TAILOR41 are summarised in Table 49. It should be noted that the rate of FN and the estimation of its costs were a key point of debate during the assessment process (see Appendix 8). Several iterations of the cost-effectiveness results for the EGFR M– patient population were produced in response to particular needs; the results presented in Table 49 are those presented at the NICE final AC meeting. The calculations are based on a rate of FN of 6.35%. This figure is derived from the subgroup of patients in TAILOR41 who received 3-weekly treatments of docetaxel (the treatment regime used in clinical practice in England and Wales).

TABLE 49 - Base-case deterministic cost-effectiveness results erlotinib vs. docetaxel second-line treatment in the EGFR M– population using evidence from TAILOR41

2L, seocnd line; Tx, treatment.

Commercial-in-confidence information has been removed.

Cost-effectiveness result	Docetaxel		Erlotinib		Incremental
Survival (mean)	Years	Months	Years	Months	Years	Months
PFS	0.409	4.91	0.287	3.45	0.122	1.46
PPS	0.731	8.77	0.641	7.70	0.089	1.07
Terminal	0.038	0.46	0.038	0.46	0.000	0.00
OS	1.178	14.13	0.967	11.60	0.211	2.53
QALYs	Undiscounted	Discounted	Undiscounted	Discounted	Undiscounted	Discounted
PFS	0.2537	0.2526	0.1853	0.1850	0.0684	0.0676
PPS	0.3459	0.3311	0.3036	0.2920	0.0423	0.0392
Terminal	0.0095	0.0092	0.0095	0.0093	0.0000	0.0001
OS	0.6091	0.5930	0.4984	0.4863	–0.1107	–0.1067
Costs	Undiscounted	Discounted	Undiscounted	Discounted	Undiscounted	Discounted
2L Tx acquisition	£342	£340	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
2L administration	£2314	£2305	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
2L Tx AEs	£585	£585	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
PFS BSC	£1531	£1524	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
PPS BSC	£5148	£4928	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
Terminal care	£3917	£3820	–	–	–	–
Total costs	£13,837	£13,504	£14,302.08	£14,049.00	£465	£545
ICER
Cost per QALY gained	–£5112 (discounted) for erlotinib vs. docetaxel (dominated)
Net benefit
£ per patient (£30,000 per QALY)	–£3746 per patient for erlotinib vs. docetaxel (dominated)

Erlotinib is dominated by docetaxel in the EGFR M– population, yielding a reduced mean survival and fewer QALYs while also involving a greater net cost of treatment. Univariate sensitivity analysis (Figure 11) for the deterministic base case indicates that the use of generic docetaxel in place of the branded product is the major factor in establishing docetaxel as the preferred option. The incidence rate of FN has a larger influence on the estimated ICER than other model parameters, but for none of model parameters is the known parameter uncertainty sufficient to alter the conclusion that erlotinib is dominated by docetaxel in the EGFR M– population. The only model input which could alter this conclusion is the incidence rate of FN in docetaxel-treated patients; this is considered in Appendix 8.

FIGURE 11.

Univariate sensitivity analysis: erlotinib vs. docetaxel second-line treatment in the EGFR M– population from TAILOR – 20 most influential parameters. DC, day case; Doc, docetaxel; Erl, erlotinib; GP, general practitioner.

A PSA (Figure 12) and the cost-effectiveness acceptability curve (Figure 13) yield a similar result: an estimated ICER of –£7709 per QALY gained, indicating that for any cost-effectiveness threshold greater than £0 per QALY there is a probability greater than 99% that erlotinib is less cost-effective than docetaxel.

FIGURE 12.

Probabilistic sensitivity analysis: scatterplot of the base-case analysis for erlotinib vs. docetaxel second-line treatment in the EGFR M– population using evidence from TAILOR. 41

FIGURE 13.

Cost-effectiveness acceptability curves for the comparison of erlotinib vs. docetaxel second-line treatment in the EGFR M– population using evidence from TAILOR. 41

Erlotinib versus best supportive care

The manufacturer of erlotinib submitted a sensitivity analysis of its main economic analysis of the population of unknown EGFR status (see next section), using survival data from a post-hoc reanalysis42 of the results of the BR.2131 trial. This analysis restricts attention to those patients who were confirmed not to have EGFR-activating mutations, that is only EGFR M– (or EGFR wild-type) disease. Inevitably the source data43 are less reliable than the main ITT analysis of BR.2131 results because of the risk of imbalance in baseline patient characteristics and the reduced sample size.

In order to replicate this sensitivity analysis, the AG has carried out a similar exercise using the same outcome data applied to the AG model structure. Figures 14 and 15 show the trajectories fitted to the trial data to populate the decision model.

FIGURE 14.

Projective models fitted to PFS data from the EGFR M– subgroup of the BR.2131 trial. WT, wild type.

FIGURE 15.

Projective models fitted to OS data from the EGFR M– subgroup of the BR.2131 trial. WT, wild type.

Deterministic results from the EGFR M– model based on subgroup EGFR M– data43 from the BR.2131 trial are summarised in Table 50. The estimated mean OS advantage of using erlotinib rather than BSC is 2.2 months, all of which occurs prior to disease progression. The corresponding gain in mean discounted QALYs is 0.116 per patient. The estimated ICER of £54,686.73 per QALY gained is above the range normally considered cost-effective. The results of univariate sensitivity analyses are summarised in Figure 16, indicating that these results are most affected by projective survival model parameters (especially for the OS model), utility model parameters and the incidence of key AEs.

TABLE 50 - Base-case deterministic cost-effectiveness results for erlotinib vs. BSC second-line treatment in the EGFR M– population (EGFR M– subgroup from the BR.21 trial31)

2L, second line; Tx, treatment.

Commercial-in-confidence information has been removed.

Cost-effectiveness result	BSC		Erlotinib		Incremental (erlotinib vs. BSC)
Survival (mean)	Years	Months	Years	Months	Years	Months
PFS	0.223	2.670	0.407	4.884	0.184	2.213
PPS	0.416	4.987	0.415	4.976	–0.001	–0.011
Terminal	0.038	0.452	0.038	0.453	0.000	0.001
OS	0.676	8.109	0.859	10.313	0.184	2.204
QALYs	Undiscounted	Discounted	Undiscounted	Discounted	Undiscounted	Discounted
PFS	0.1414	0.1413	0.2585	0.2575	0.1171	0.1163
PPS	0.1967	0.1911	0.1963	0.1911	–0.0005	0.0001
Terminal	0.0094	0.0093	0.0094	0.0093	0.0001	0.0000
OS	0.3475	0.3416	0.4641	0.4579	0.1167	0.1163
Costs	Undiscounted	Discounted	Undiscounted	Discounted	Undiscounted	Discounted
2L Tx acquisition	£0.00	£0.00	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
2L administration	£0.00	£0.00	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
2L Tx AEs	£533.70	£533.31	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
PFS BSC	£827.93	£827.33	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
PPS BSC	£2961.94	£2878.02	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
Terminal care	£3882.90	£3836.73	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
ICER
Total costs	£8206.46	£8075.39	£14,596.93	£14,436.92	£6390.47	£6361.53
Cost per QALY gained	£54,686.73 for erlotinib vs. BSC

FIGURE 16.

Univariate sensitivity analysis: erlotinib vs. BSC second-line treatment in the EGFR M– population subgroup of the BR.2131 trial – 20 most influential parameters. Erl, erlotinib.

Probabilistic sensitivity analysis incorporating uncertainty in all model parameters indicates a slightly lower estimated ICER of £54,184 per QALY gained. Examination of the PSA scatterplot (Figure 17) and the cost-effectiveness acceptability curves (Figure 18) indicate strong general confidence that erlotinib exhibits a high ICER when compared with BSC in this subgroup (0% of simulations favour erlotinib at a willingness-to-pay threshold of £30,000 per QALY gained, and 12% at £50,000 per QALY gained).

FIGURE 17.

Probabilistic sensitivity analysis: scatterplot of the base-case cost-effectiveness analysis for erlotinib vs. BSC second-line treatment in the EGFR M– subgroup of the BR.2131 trial.

FIGURE 18.

Cost-effectiveness acceptability curves for the comparison of erlotinib vs. BSC second-line treatment in the EGFR M– subgroup from the BR.2131 trial.

Epidermal growth factor mutation status unknown population

Deterministic results from the EGFR status unknown model based on data from the BR.2131 trial are summarised in Table 51. The estimated survival advantage of using erlotinib rather than BSC is 2.1 months, of which 1.7 months is prior to disease progression. The corresponding gain in mean discounted QALYs is 0.103 per patient. The overall incremental cost per patient is higher for erlotinib use (£6314 discounted), primarily because of the acquisition cost of erlotinib (£5677 discounted). The estimated ICER of £61,132 per QALY gained is well beyond the range normally considered cost-effective. The results of univariate sensitivity analyses are summarised in Figure 19, indicating that these results are unaffected by uncertainty in almost all model parameters. The only exceptions are the intercept parameter value in the Nafees et al. 75 utility model (i.e. the baseline NSCLC population utility value in patients with stable disease) and the incidence of FN when docetaxel is used.

TABLE 51 - Base-case deterministic cost-effectiveness results for erlotinib vs. BSC second-line treatment in the EGFR unknown population using evidence from BR.2131 trial

2L, second line; Tx, treatment.

Commercial-in-confidence information has been removed.

Cost-effectiveness result	BSC		Erlotinib		Incremental (erlotinib vs. BSC)
Survival (mean)	Years	Months	Years	Months	Years	Months
PFS	0.235	2.815	0.374	4.490	0.140	1.675
PPS	0.403	4.831	0.439	5.267	0.036	0.435
Terminal	0.038	0.458	0.038	0.454	0.000	–0.004
OS	0.675	8.104	0.851	10.211	0.176	2.106
QALYs	Undiscounted	Discounted	Undiscounted	Discounted	Undiscounted	Discounted
PFS	0.1490	0.1488	0.2376	0.2369	0.0886	0.0881
PPS	0.1906	0.1869	0.2077	0.2023	0.0171	0.0153
Terminal	0.0095	0.0094	0.0094	0.0093	–0.0001	–0.0001
OS	0.3491	0.3452	0.4548	0.4484	0.1057	0.1033
Costs	Undiscounted	Discounted	Undiscounted	Discounted	Undiscounted	Discounted
2L Tx acquisition	£0.00	£0.00	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
2L administration	£0.00	£0.00	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
2L Tx AEs	£562.53	£561.77	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
PFS BSC	£873.25	£872.06	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
PPS BSC	£2853.14	£2798.84	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
Terminal care	£3938.22	£3900.12	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed	Commercial-in-confidence information has been removed
Total costs	£8227.15	£8132.79	£14,610.64	£14,446.38	£6383.49	£6313.59
ICER
Cost per QALY gained	£61,161.81 for erlotinib vs. BSC

FIGURE 19.

Univariate sensitivity analysis: erlotinib vs. BSC second-line treatment in the EGFR unknown subgroup of the BR.2131 trial – 20 most influential parameters. Erl, erlotinib.

Probabilistic sensitivity analysis incorporating uncertainty in all model parameters indicates a slightly lower estimated ICER of £59,973 per QALY gained. Examination of the PSA scatterplot (Figure 20), and the cost-effectiveness acceptability curves (Figure 21) indicate strong general confidence that erlotinib is not more cost-effective than BSC in this population (0% of simulations favour erlotinib at a willingness-to-pay threshold of £30,000 per QALY gained).

FIGURE 20.

Probabilistic sensitivity analysis: scatterplot of the base-case cost-effectiveness analysis for erlotinib vs. BSC second-line treatment in the EGFR unknown population from the BR.2131 trial.

FIGURE 21.

Cost-effectiveness acceptability curves for the base case for erlotinib vs. BSC second-line treatment of NSCLC in the EGFR unknown population using results from the BR.2131 trial.

Summary and discussion of Assessment Group model results

The very weak evidence base for comparative second-line treatments, especially in subgroups defined by EGFR-TKI activating mutation status, has severely restricted the AG’s ability to assess the relative cost-effectiveness of all potential treatments and comparators indicated in appraisal scope.

In the absence of reliable RCT data comparing second-line treatments in a population with confirmed EGFR-activating mutations, no cost-effectiveness analysis could be undertaken. This is a serious information deficit that urgently requires remedy. In particular, this problem prevents any consideration of gefitinib as a potential post-progression treatment, as gefitinib is licensed for use only in patients with activating mutations. The AG is aware that current treatments for patients who have EGFR M+ disease are evolving and include the use of platinum-doublet chemotherapy after progression following EGFR-TKI treatments; however, no robust data are available for use in this appraisal.

TAILOR41 comparing the effectiveness of docetaxel monotherapy and erlotinib is the only RCT with data currently available in a population with confirmed disease and lacking EGFR-activating mutations. Cost-effectiveness analysis using data from this trial indicates that erlotinib is dominated by docetaxel in the EGFR M– population yielding a reduced mean survival and fewer QALYs while involving a greater net cost of treatment. When additional studies are published for the EGFR M– population, it will become clearer whether this result is confirmed or brought into question.

A significant survival benefit for docetaxel may be translated into good cost-effectiveness over erlotinib (erlotinib is dominated by docetaxel, ICER = –£5112 per QALY gained), on the basis that generic docetaxel is priced at the level corresponding to that currently paid by the NHS. If published list prices are substituted, docetaxel looks much less attractive.

A subgroup analysis of the BR.2131 trial comparing erlotinib with BSC in those patients without EGFR-activating mutations confirms that erlotinib generates survival advantages, but at high cost, so that the estimated ICER is high for the EGFR M– population (£54,687 per QALY gained).

In the case of patients who are eligible for second-line therapy but for whom definitive determination of EGFR mutation status is not available for any reason, cost-effectiveness analysis based on the whole of the BR.2131 trial cohort also yields a high ICER value for the EGFR unknown population (£61,132 per QALY gained).

Thus, on the basis of the clinical-effectiveness data currently useable for economic analysis, it does not appear that second-line erlotinib for NSCLC is an attractive option in the EFGR M– or EGFR unknown populations, and at present there are no sources of effectiveness data on which to base an assessment of erlotinib compared with any other option in those patients with confirmed EGFR-activating mutations. The absence of suitable head-to-head trials in the era of EGFR mutation testing is therefore the main limitation on the economic analyses that could be carried out by the AG.

The analyses described here do not take into account the issue of patient experience and preferences in the delivery of second-line treatment, in particular, that oral therapy is widely preferred by patients and clinicians to treatments delivered intravenously. This affects only the comparison made between erlotinib and docetaxel in the EGFR M– population. One possible approach to dealing with this concern is to include an additional utility increment applied only to erlotinib in the analysis to represent the reduction in pain, anxiety and disruption to everyday activities from switching to an oral treatment. There is no objective way to measure such an effect at present. However, a sensitivity analysis can be carried out by assessing the effect of the maximum possible patient health utility increment on the estimated ICER. This is achieved by setting the additional increment at the level which corresponds to returning a patient to the average QoL experienced in the general population at the equivalent mean age (utility about 0.8). This requires raising the EQ-5D score by 0.155, which reduces the incremental loss of QALYs slightly but leaves erlotinib still dominated by docetaxel. This extreme sensitivity analysis indicates that any realistic assessment of utility advantage due to oral therapy is very unlikely to alter the relative cost-effectiveness of erlotinib and docetaxel in the EGFR M– population.

Assessment of factors relevant to the NHS and other parties

This review has highlighted that a key development since TA16221 in 2009 has been the expiration of the patent for docetaxel. This means that generic versions of docetaxel are now available in England and Wales at a substantially reduced cost to the NHS. In TA162,21 NICE recommends the use of docetaxel and erlotinib as second-line treatments for patients with NSCLC. Erlotinib is currently recommended only on the basis that it is provided by the manufacturer at an overall treatment cost equal to that of docetaxel. Docetaxel is now available at 10% of its original list price. Clearly, this reduced price of docetaxel has resource implications that are relevant to the NHS, NICE and the manufacturer of erlotinib. In particular, the results of the AG’s cost-effectiveness analysis comparing erlotinib with docetaxel show that docetaxel is more cost-effective than erlotinib in an EGFR M– patient population.

Recent advances in lung cancer diagnosis and treatments have revealed that expected clinical benefit from available lung cancer treatments can be positively or negatively affected by a patient’s EGFR mutation status. The AG therefore considers it imperative that EGFR mutation tests are routinely available for all NSCLC patients at the time of diagnosis, prior to treatment. The NHS is making every effort to offer timely EGFR mutation tests to patients with NSCLC across England and Wales, however clinical expert opinion is that EGFR mutation tests are not currently routinely available in all centres because of the unavailability of testing facilities and inconclusive results.

In patient populations in which docetaxel is preferred to erlotinib from a cost-effectiveness perspective, there are concerns that this represents a backwards step in patient treatment options. Docetaxel is administered as an i.v. infusion, which means that patients are required to attend hospital as a day-case to receive this treatment. Replacing erlotinib (oral therapy) with docetaxel (i.v. therapy) has major implications not only for NHS resource use and staff, but also in terms of patient preference.

Clinical effectiveness results

Cost-effectiveness results

The AG developed a de novo economic model for the specific purpose of this Multiple Technology Appraisal and carried out several cost-effectiveness analyses.

For the EGFR M+ population, the AG was not able to carry out a cost-effectiveness analysis of available treatments, as there is an absence of relevant direct clinical trial evidence in this patient population.

For the EGFR M– population, the AG compared docetaxel with erlotinib using data from TAILOR. 41 In this comparison erlotinib was dominated by docetaxel, yielding a reduced mean survival and fewer QALYs while also involving a greater net cost of treatment. Docetaxel yielded a survival advantage over erlotinib of 2.5 months. The overall treatment cost of docetaxel was £545 lower than the cost of erlotinib. The AG estimated the size of the erlotinib versus docetaxel ICER to be –£5112 per QALY gained. However, if published list prices are used instead of eMit prices, the ICER for docetaxel versus erlotinib increases to over £57,000 per QALY gained, with a probabilistic ICER of nearly £70,000 per QALY gained.

For the EGFR M– population, the AG also compared erlotinib versus BSC in a sensitivity analysis using data from the post-hoc reanalysis of BR.2143 described in the manufacturer’s submission submitted by Roche (UK) Ltd. In this comparison, erlotinib yielded a survival advantage over BSC of 2.2 months, with an incremental QALY gain of 0.116. The overall treatment cost of erlotinib was £6362 higher than the cost of BSC. The AG estimated the size of the erlotinib versus BSC ICER to be £54,687 per QALY gained. This ICER is above the range normally accepted to be cost-effective. PSA incorporating uncertainty in all model parameters indicates a slightly lower ICER of £54,984 per QALY gained.

For the EGFR unknown population, the AG compared erlotinib and BSC using data from the BR.2131 trial. In this comparison, erlotinib yielded a survival advantage of 2.1 months, with an incremental QALY gain of 0.103. The overall treatment cost of erlotinib was £6312 higher than the cost of BSC. The AG estimated the size of the erlotinib versus BSC ICER to be £61,132 per QALY gained. This ICER is outside the range normally accepted to be cost-effective. PSA incorporating uncertainty in all model parameters indicates a slightly lower ICER of £59,973 per QALY gained.

Search strategy

1. randomized controlled trial.pt.

2. controlled clinical trial.pt.

3. randomized.ab.

4. placebo.ab.

5. randomly.ab.

6. trial.ab.

7. or/1-6

8. (animals not (humans and animals)).sh.

9. 7 not 8

10. exp Carcinoma, Non-Small-Cell Lung/ or nsclc.ti,ab.

11. (non-small or non small or nonsmall).ti,ab.

12. (lung or pulmonary or bronchus or bronchogenic or bronchial or bronchoalveolar or alveolar).ti,ab.

13. (neoplasm$ or cancer$ or carcinoma$ or adenocarcinoma$ or angiosarcoma$ or chrondosarcoma$ or sarcoma$ or teratoma$ or lymphoma$ or blastoma$ or microcytic$ or carcinogenesis or tumour$ or tumor$ or metast$).ti,ab.

14. 10 or (and/11-13)

15. (erlotinib or tarceva or “osi 774”).ti,ab.

16. (gefitinib or iressa or ZD 1839).ti,ab.

17. 15 or 16

18. 9 and 14 and 17

19. limit 18 to English language

Search strategy

1. Randomized Controlled Trial/

2. Randomization/

3. Single blind procedure/

4. Double blind procedure/

5. Double blind procedure/

6. Crossover procedure/

7. Randomi?ed controlled trial$.tw.

8. random$.ti,ab.

9. placebo.ti,ab.

10. or/1-9

11. animal/ not (animal/ and human/)

12. 10 not 11

13. exp lung non small cell cancer/ or nsclc.ti,ab.

14. (non-small or non small or nonsmall).ti,ab.

15. (lung or pulmonary or bronchus or bronchogenic or bronchial or bronchoalveolar or alveolar).ti,ab.

16. (neoplasm$ or cancer$ or carcinoma$ or adenocarcinoma$ or angiosarcoma$ or chrondosarcoma$ or sarcoma$ or teratoma$ or lymphoma$ or blastoma$ or microcytic$ or carcinogenesis or tumour$ or tumor$ or metast$).ti,ab.

17. 13 or (and/14-16)

18. exp erlotinib/

19. (erlotinib or tarceva or “osi 774”).ti,ab.

20. exp gefitinib/

21. (gefitinib or iressa or ZD 1839).ti,ab.

22. or/18-21

23. 12 and 17 and 22

24. limit 23 to English language

Search strategy

#1. MeSH descriptor: [Carcinoma, Non-Small-Cell Lung] explode all trees

#2. “non-small-cell lung cancer”:ti,ab,kw (word variations have been searched)

#3. erlotinib or tarceva:ti,ab,kw (word variations have been searched)

#4. gefitinib or iressa:ti,ab,kw (word variations have been searched)

#5. #1 or #2

#6. #3 or #4

#7. #5 and #6

Search strategy

((erlotinib or tarceva or gefitinib or iressa)) AND lung cancer