Clinical effectiveness and cost-effectiveness of first-line chemotherapy for adult patients with locally advanced or metastatic non-small cell lung cancer: a systematic review and economic evaluation

Search strategy

Three electronic databases (MEDLINE, EMBASE and The Cochrane Library) were searched from January 1990 to August 2010 for randomised controlled trials (RCTs), systematic reviews and economic evaluations.

Patient populations

Chemotherapy-naive adult patients with locally advanced or metastatic NSCLC were included.

Interventions and comparators

Studies that compared any first-line chemotherapy currently licensed in Europe and recommended by NICE were considered.

Outcomes

Data on any of the following outcomes were included in the assessment of clinical effectiveness: overall survival (OS), OS at 1 and 2 years, progression-free survival (PFS), time to progression (TTP), tumour overall response rate, quality of life (QoL) and adverse events (AEs). For the assessment of cost-effectiveness, outcomes included incremental cost per life-year gained and incremental cost per quality-adjusted life-year (QALY) gained.

Application of inclusion/exclusion criteria

Two reviewers independently screened all titles and/or abstracts including economic evaluations. The full manuscript of any publication judged to be relevant by a reviewer was obtained and assessed for inclusion or exclusion. Two reviewers assessed the relevance of each publication; any discrepancies were resolved by consensus and, where necessary, a third reviewer was consulted.

Data extraction and quality assessment

Data were extracted into a Microsoft Access 2007 database (Microsoft Corporation, Redmond, WA, USA). All trials were assessed for methodological quality using criteria based on the Centre for Reviews and Dissemination guidance. The results of clinical and economic data extraction and quality assessment are summarised in the tables and narrative description.

Evidence synthesis

Where appropriate, relative treatment effects for OS, PFS, TTP and survival risk at years 1 and 2 were estimated using a standard meta-analysis for head-to-head comparisons between interventions based on intention-to-treat analyses. Mixed-treatment comparison methodology was used for the clinical effectiveness outcomes of OS, PFS, TTP and survival risk at 1 and 2 years.

Quality assessment

Overall, the quality of the included RCTs was poorer than expected: there were few trials with fully reported methods and the definitions of the health outcomes used often differed between trials.

Clinical effectiveness review: efficacy data

All 23 clinical trials were published between 2001 and 2010 and included a total of 11,428 randomised patients. Of the 20 multicentre trials, six were international; the three single-centre trials were based in Taiwan. Seventeen trials were assessed as being sufficiently powered to evaluate OS. Median follow-up of patients ranged from 8 to 45 months. Doses of chemotherapy drugs varied, median number of chemotherapy cycles ranged from 2.6 to 6 and chemotherapy treaments were administered either by intravenous (i.v.) infusion or orally.

When the three GEF trials were compared with the other included trials, the proportion of males to females was much less; the percentage of males in the GEF trials ranged from 21% to 37%. These three trials were conducted in East Asian countries and had somewhat different patient populations compared with the other trials. Two of these trials included only patients with EGFR M+ tumour status, and one trial included patients with pulmonary adenocarcinoma who were never-smokers or were former light smokers.

Twenty-three trials were included within the network of trials for the clinical analysis. The direct evidence for the NSCLC population with squamous disease included 18 trials (> 7000 patients and > 6000 deaths). These same 18 trials plus subgroup data from an additional two studies were included in the analysis of the NSCLC population with non-squamous disease. Participants of three studies, conducted entirely within East Asian countries, constituted the EGFR M+ NSCLC population. In general, there was consistency between the results of the direct meta-analyses and the mixed-treatment comparison analyses, and also very good consistency across individual trials in the within-group comparisons.

Among NSCLC patients with squamous disease, there were no statistically significant differences between any of the four chemotherapy regimens (DOC + PLAT, GEM + PLAT, PAX + PLAT, VNB + PLAT) in terms of increasing OS. However, both the direct and indirect evidence suggests a potential non-statistically significant advantage in terms of OS for GEM + PLAT [direct meta-analysis 1: hazard ratio (HR) = 1.08; 95% confidence interval (CI) 0.98 to 1.20] and for DOC + PLAT (direct meta-analysis 1: HR = 0.89; 95% CI 0.78 to 1.00; mixed-treatment comparison 1, HR = 0.92; 95% CI 0.81 to 1.03) compared with VNB + PLAT. Analyses of 1- and 2-year survival support this conclusion.

For patients with non-squamous NSCLC there is borderline statistically significant evidence to suggest that PEM + PLAT increases OS compared with GEM + PLAT (direct meta-analysis 1, HR = 0.85; 95% CI 0.73 to 1.00). However, there is no statistically significant evidence to suggest that PEM + PLAT compared with GEM + PLAT increases PFS (mixed-treatment comparison 1, HR = 0.85; 95% CI 0.74 to 0.98).

Among patients with EGFR M+ status, OS was not statistically significantly different in those treated with GEF and those receiving PAX + PLAT or in those treated with GEF compared with those treated with DOC + PLAT. There was a statistically significant improvement in PFS among those patients treated with GEF compared with those treated with DOC + PLAT or PAX + PLAT. However, there was significant quantitative heterogeneity between the two trials comparing GEF with PAX + PLAT, which requires further exploration.

It remains unknown whether or not the clinical effectiveness of PEM + PLAT is superior to that of GEF monotherapy for patients with non-squamous disease. The relative clinical effectiveness of PEM + PLAT in patients who are EGFR M+ is unknown.

Clinical effectiveness review: adverse events

Across all the chemotherapy arms of the included trials, the most common AEs were neutropenia, anaemia and leucopenia. Rates of haematological AEs were similar for all the chemotherapy drugs with the exception of GEF, which appears to be associated with a significantly lower severe AE rate than some of the other drugs. The trials often varied in the way that AEs were defined, measured and reported.

Clinical effectiveness review: quality of life

Twelve trials reported QoL outcomes using a variety of instruments/tools. Seven trials reported no significant difference in QoL and four trials reported some significant differences between treatment groups. A lack of reporting of QoL data is a feature of the great majority of trials assessing outcomes of treatment for patients with NSCLC. This, despite its relevance to patients and clinicians, is a major shortcoming of lung cancer research. Measuring QoL outcomes in patients with advanced NSCLC is difficult mainly because of the severity of symptoms, the side effects of chemotherapy and early deaths associated with NSCLC. However, the British Thoracic Oncology Group Trial 2 has shown that it is feasible to collect QoL data in patients with performance status (PS) 0–2, stage IIIB/IV NSCLC disease within a clinical trial setting.

Cost-effectiveness review: summary

None of the seven included studies were directly relevant to decision-making in the NHS because they are not UK focused and/or they do not estimate incremental cost-effectiveness ratios (ICERs) in terms of cost per QALY gained.

Economic results: patients with squamous disease

The four third-generation chemotherapy agents, when used in combination with PLAT for first-line treatment of advanced or metastatic NSCLC, are often considered to exhibit similar effectiveness, when compared in terms of standard statistical measures (e.g. p-values). However, the mixed-treatment comparison analysis undertaken by the AG which informs the current model does indicate important differences which, when combined with differences in the management of the condition and acquisition cost, provide a basis for differentiating between treatment options and arriving at some robust conclusions:

• In both deterministic and probabilistic analyses for both the base-case and alternative pricing scenarios, VNB doublets yield the least patient benefit (as measured by expected discounted QALYs), and are not the least expensive option. As a result, VNB cannot be considered to provide either optimal effective or cost-effective chemotherapy treatment.

• PAX doublets are consistently minimum cost options and therefore represent the initial ‘good value’ treatment, to be supplanted only if an alternative option yields greater benefit at an acceptable ‘willingness-to-pay’ (WTP) threshold.

• The choice of preferred alternative main agent to PAX generally favours DOC over GEM as its greater effectiveness appears to outweigh the additional acquisition cost, although both lie on the efficiency frontier.

Economic results: patients with non-squamous disease

The addition of a PEM doublet to the four third-generation chemotherapy agents changes the relationship between the regimens, because of the clear outcome advantage of PEM therapy in terms of the improved expected survival of patients with non-squamous disease. However, the high price of branded PEM compared with the other drugs (in most cases available generically) means that PEM is preferred on cost-effectiveness grounds only if the WTP threshold is set > £37,000 per QALY (or £50,000 per QALY if eMIT prices are assumed). This means that PAX remains a viable treatment (and possibly GEM and DOC). However, VNB is clearly not cost-effective in either scenario.

Economic results: patients who are epidermal growth factor receptor mutation positive

The base-case analyses for GEF compared with the two chemotherapy doublets (PAX and DOC) for which evidence is available show poor cost-effectiveness for GEF. Results are improved somewhat by disaggregating the three GEF trials, but even then cost-effective ICERs (< £30,000 per QALY gained) are obtained only for the second alternative scenario [Western Japan Thoracic Oncology Group (WJTOG) trial only] based on the smallest RCT comparing GEF with the DOC + CIS doublet.

Limitations

The limitations of the report can be summarised as follows: very few trials reported QoL data; AEs from the different trials were difficult to compare; CARB and CIS were treated as being similarly effective in the clinical analyses; and owing to the large volumes of data available for patients with lung cancer, the methods employed in the review do not always match the methods stated in the original protocol. Finally, the quality of the included trials was poorer than anticipated and this finding must be taken into consideration when interpreting the results of the clinical and economic analyses presented.

Research recommendations

The design of future lung cancer trials needs to reflect the influence of factors such as histology, genetics and the new prognostic biomarkers that are currently being identified. In addition, trials will need to be adequately powered so as to be able to test for statistically significant clinical effectiveness differences within patient populations. New initiatives are in place to record detailed information on the precise chemotherapy (and targeted chemotherapy) regimens being used, together with data on age, cell type, stage of disease and PS, allowing for very detailed observational audits of management and outcomes at a population level. It would be useful if these initiatives could be expanded to include the collection of health economics data.

Implications for practice

Closer examination of clinical effectiveness and cost-effectiveness data means that we have been able to provide a comprehensive framework of information for three subpopulations of patients with NSCLC that clinicians can refer to as they attempt to balance patient factors, available treatments, treatment costs and AEs in their daily decision-making.

Concluding remarks

The completion of this review has taken a significant length of time and during that period there has been explicit acknowledgement in the published literature of the important differences in the characteristics of patients who previously were identified as having NSCLC. It is anticipated that no further RCTs will be carried out involving patients with NSCLC as a homogeneous group, but that consideration of the important patient subgroups will take precedence and allow for the development of more specialised and targeted treatments which, in turn, will require RCTs of increasingly sophisticated design.

Incidence and prevalence

Lung cancer is the most common cancer in the world and the second most common cancer diagnosed in the UK after breast cancer. In 2008, 40,806 new cases were diagnosed in the UK: 32,546 in England and 2403 in Wales. 1 Lung cancer is rarely diagnosed in people aged < 40 years and 86% of cases occur in people aged > 60 years. 1 Table 1 provides an overview of lung cancer statistics in the UK. The European age-standardised incidence rate of lung cancer in 2008 was 45.6 per 100,000 population in England and 52.2 per 100,000 population in Wales. 1 The UK incidence rate in males is similar to incidence rates in most of Western Europe and lower than those in most of Eastern Europe. The UK incidence rate in females is one of the highest rates in the European Union (EU). 1 There is an increased incidence of lung cancer in individuals from the lowest socioeconomic strata. 2–4 In 2008, around 65,000 individuals were living with lung cancer in the UK,1 the majority of them male. 1

Causation

Smoking causes around 90% of lung cancer deaths in men and > 80% of lung cancer deaths in women in the UK. 5 Other causes include radon exposure, air pollution, heredity and occupational exposures such as asbestos and industrial chemicals. 6

Survival

There were 35,261 lung cancer-related deaths in the UK in 2008. 1 Prognosis is very poor; lung cancer is usually asymptomatic in the early stages and two-thirds of patients are diagnosed at a late stage when curative treatment is not possible. Twenty-seven per cent of male and 30% of female lung cancer patients in England and Wales survive for 1 year; 7% and 9%, respectively, survive 5 years. 1 According to the National Lung Cancer Data Audit (LUCADA) 2006–8, the median survival for individuals with lung cancer in England is 203 days (interquartile range 62–545 days). 7

There are many factors that affect lung cancer survival rates, including smoking status, general health, sex, race and cancer treatments. For example, survival rates at 1 and 3 years are significantly higher among Asian than white lung cancer patients, regardless of age. 1

Disease staging

The stage of lung cancer at diagnosis reflects the degree of spread of cancer and is crucially important to determine which patients have potentially curative disease, and which do not, and this helps to define a patient's prognosis. TNM (tumour, node and metastasis) classification provides a system for staging the extent of cancer. Table 2 shows the seventh edition of the Union for International Cancer Control (UICC) TNM system9 for classification of NSCLC disease stage. T refers to the size of the primary tumour, N refers to the involvement of the lymph nodes and M refers to the presence of metastases or distant spread of the disease. It should be noted that all of the trial evidence in this review would have used the UICC sixth edition (or lower), as the seventh edition has only been implemented in the UK since January 2010. Table 2 compares the stage from the sixth edition, which has been modified, with the new stage in the seventh edition. Table 3 shows the surgical stage groupings in the seventh TNM classification.

Performance status

Performance status is used to quantify cancer patients' general well-being and may be used to determine whether or not a patient is fit enough to receive chemotherapy, whether or not a chemotherapy dose adjustment is necessary, and to quantify how much supportive care a patient may require. There are three main scales used to measure PS: the World Health Organization (WHO) PS scale,10 the Karnofsky Performance Status (KPS) scale10 and the Eastern Cooperative Oncology Group (ECOG) PS scale. 11 A summary of the WHO PS scale is shown in Table 4 as this is the most commonly used scale in clinical practice in the UK. 10 A score of 0 on the WHO scale indicates a patient is completely able to look after him/herself and a score of 4 indicates that a patient requires a lot of support.

Histology

Non-small cell lung cancer accounts for approximately 84% of all lung cancers diagnosed and the remaining 16% are small cell lung cancers. The main subtypes of NSCLC are squamous cell carcinoma (33%) and non-squamous cell carcinoma (29%); the latter is made up of adenocarcinoma (25%) and large cell carcinoma (4%). Approximately 36% of patients are listed as having NSCLC ‘not-otherwise specified’ and 1% as having carcinoma in situ. 12

Squamous cell carcinoma commonly begins in the bronchi, centrally in the lungs. Adenocarcinoma starts in the periphery of the lungs and can tend to be present for a long time before it is detected. It is the type of lung cancer usually found in non-smokers and is the most common type seen in women. Large cell carcinomas often occur in the outer regions of the lungs, and tend to grow rapidly and spread more quickly than some other forms of NSCLC. 13

Histological confirmation (i.e. a diagnosis made by taking a sample of tissue or cells) is an important element of diagnosis because it helps to determine a patient's treatment pathway. However, it is noted that histological confirmation is not always straightforward and there are several key issues that must be noted. For example, tumour heterogeneity in the context of small histological or cytological samples size, interobserver variation, the absence of centralised pathology review, and the lack of any tested biological hypothesis which explains this observation that some drugs do better when the term squamous is applied to the biopsy and some when it is not.

Despite these cautions, more and more treatments are being recommended for different types of patients as recent evidence suggests that newer drugs are beneficial in certain histological subtypes of NSCLC. For example, pemetrexed (Alimta^®, Eli Lilly and Company; PEM) is beneficial in patients with non-squamous carcinoma. 13 (Note: PEM is licensed and recommended for use in patients with adenocarcinoma and large cell carcinoma; however, for the purposes of this report we refer to the use of PEM in patients with non-squamous disease.) A significant proportion of patients are diagnosed based on clinical examination and radiological investigations alone, without histological evidence. According to LUCADA, in England and Wales, histological confirmation of the cancer diagnosis is made in 72% of cases, although there is wide (regional) variation from 25% to > 85%. 14 Given that more chemotherapy options are becoming available which alter the potential treatment pathway of a patient, histological testing is expected to be carried out more frequently and in a more standardised way in the future. Recent NICE guidance for the first-line treatment of NSCLC recommends histological testing and, therefore, histological testing rates are expected to increase. 7 The proportion of patients diagnosed as having disease ‘not otherwise specified’ will decrease in time [and the proportion of patients with non-squamous and EGFR mutation-positive (M+) disease will gradually rise], although regional variation will inevitably continue.

Epidermal growth factor receptor mutation status

Improvements in the understanding of the molecular and biological basis of lung cancer have led to the identification of a number of drugs that target proteins on cancer cells for the treatment of lung cancer. Among the most studied is EGFR tyrosine kinase inhibitors (TKIs), which target proteins on cancer cells and are an effective treatment for patients with tumours with activating mutations of the epidermal growth factor receptor-tyrosine kinase (EGFR-TK)15 (EGFR M+) and are often referred to as part of a group of treatments labelled ‘targeted chemotherapy’. Analysis of predictive tumour markers is necessary to identify patients with EGFR M+ who would then be candidates for such targeted treatment. 16

Clinical consensus is that EGFR M+ status will be present in around 10% of patients within the overall population. A study by Rossell et al. 17 found that 17% of Spanish patients with non-squamous NSCLC were EGFR M+. There are various clinical and lifestyle factors associated with the likelihood of the presence of EGFR mutation; for example, the rate of EGFR M+ status is higher among East Asian female non-smokers with adenocarcinoma than among white British male smokers with squamous cell carcinoma. EGFR mutation status can act as both a predictor of response to chemotherapy treatment (identification of subgroups of populations that would benefit from EGFR-targeted therapy) and a prognostic factor (indicator or the likely natural course of the disease).

Treatment options for non-small cell lung cancer

It would be useful to define chemotherapy options before discussing treatment options in detail. Chemotherapy is the treatment of cancer using chemical substances. Chemotherapy drugs work to destroy cancer cells by preventing them from multiplying. Treatment consists of either a chemotherapeutic agent or a molecularly targeted agent such as EGFR. Chemotherapies are generally non-specific in cellular action; they preferentially target rapidly proliferating cells and do not discriminate between malignant and non-malignant cells.

Figure 1 shows a treatment pathway for patients with NSCLC and shows estimates of the proportions and numbers of patients with NSCLC along the treatment pathway in England and Wales based on histology and staging data, NICE guidelines7 and NICE guidance. 26,27 Recommendations for patients with small cell lung cancer are not discussed in this report.

FIGURE 1.

Treatment pathway for patients with NSCLC. BSC, best supportive care; PLAT, platinum.

Thirty per cent of patients with NSCLC are diagnosed with stage I–IIIA disease (personal communication with Dr Michael Peake, Glenfield Hospital, using unpublished LUCADA data from 2009). These patients are suitable for potentially curative surgery or radical radiotherapy. Surgery for NSCLC consists of lobectomy, pneumonectomy and wedge resection. Approximately 50% of patients undergoing these procedures will relapse and will then be eligible for further treatment. 18 Patients with stage IIIA–IIIB disease who are not amenable to surgery can be treated with potentially curative chemoradiation.

Seventy per cent of patients with NSCLC have stage IIIB or IV disease and a PS of 0 or 1 at the time of diagnosis. These patients are assessed for their suitability for first-line chemotherapy; less than half (48%) of patients who are assessed actually receive it. 8 Among those who receive chemotherapy, almost half will respond to treatment and have either a complete or a partial response. Of these patients, a relatively small proportion can go on to have maintenance treatment and only 28% are suitable for second-line chemotherapy. 23

The majority of patients with NSCLC are diagnosed late and have metastatic or locally advanced disease. Therefore, up to 50% of patients are treated with best supportive care (BSC) alone. During all stages of treatment, patients receive BSC or ‘active supportive care’ in addition to any anticancer treatment. In the recently published lung cancer guidelines,7 NICE defines ‘supportive care’ as ‘the multidisciplinary holistic care offered to all patients and their carers throughout the pathway to help them cope with cancer and treatment of it. Best supportive care packages include options for information giving, symptom control and psychological, social and spiritual support. Palliative care provides a similar holistic approach, but is specific to those patients with advanced progressive illness’ (p. 98). 7

First-line treatment options for patients with NSCLC are shown in Figure 2. Less than 70% of patients with NSCLC have stage IIIB or stage IV disease, which equates to 20,433 patients. The percentage of patients in each stage is from 2009 audit data (M Peake, personal communication). The proportion of patients receiving BSC, chemotherapy, radiotherapy and surgery, stratified by stage is derived from LUCADA data for 2009 (personal communication with Dr Paul Beckett, Queens Hospital, using unpublished LUCADA data from 2009). These proportions have been applied to the most up-to-date incidence rates for NSCLC in England and Wales. 24 It should be noted that disease stage was recorded in 81% of cases and that these cases represent 98% of expected incidence cases for 2009; therefore, as a result of these missing data the total percentage of patients receiving BSC, radiotherapy, chemotherapy and surgery ranges from 70% to 94% within each disease stage (and does not equal 100%). In addition, the percentage of people receiving radiotherapy includes both those receiving radical and those receiving palliative treatment.

FIGURE 2.

Treatment options for patients with NSCLC in the first-line setting. CTX, chemotherapy; RT, radiotherapy; SX, surgery.

Outcome measures

Survival is considered the most reliable cancer end point within a randomised controlled trial (RCT), and when trials can be conducted to adequately assess survival it is usually the preferred end point. Overall survival (OS) is measured as the time from randomisation to death from any cause; median survival is the point in time at which 50% of people with a condition will have died and 50% are still alive. Year-1 and -2 survival risks are defined as the probability of survival in intervals of time elapsed from randomisation to years 1 and 2, respectively.

The majority of trials also report progression-free survival (PFS) as an intermediate surrogate measure of survival. PFS measures the length of time between randomisation until tumour progression or death from any cause; unlike OS, PFS is not an unequivocal outcome measure and is often determined by how frequently patients are monitored.

Tumour progression is defined as at least a 20% growth in the size of the tumour or spread of the tumour since the beginning of treatment. 28 Time to progression (TTP) is defined as the time from randomisation until tumour progression (and does not include death). The majority of RCTs also measure overall response rate (ORR), which is the proportion of people who show a response (the tumour shrinks), which can be complete or partial. Stable disease is recorded when there is no response and the tumour does not change in size. Stable disease also means that no new tumours have developed and that the cancer has not spread to any new regions of the body. 28

Adverse event (AE) and quality-of-life (QoL) data are also measures of important clinical benefit and provide information on how well chemotherapy is tolerated. In patients with advanced NSCLC, palliative chemotherapy is given to improve QoL. The EuroQol 5D (European Quality of Life-5 Dimensions; EQ-5D) is a standardised generic instrument for measuring health-related quality of life (HRQoL). It provides a utility score for health and a self-rating of HRQoL. Other commonly used QoL tools within NSCLC trials are the European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire (EORTC QLQ)-C3029 and the lung cancer-specific module QLQ-LC13,30 the Lung Cancer Symptom Scale (LCSS)31 and the Functional Assessment of Cancer Therapy – Lung (FACT-L) questionnaire. 32 Both AEs resulting from the disease itself and those due to chemotherapy have a considerable impact on HRQoL. 33

Despite QoL being both a vitally important measure of a patient's general emotional, physical and mental well-being and a very relevant measure of the ‘success’ of chemotherapy treatment primarily because advanced stage NSCLC is not curable, a minority of trials address QoL issues. When QoL has been examined, patients receiving chemotherapy report better scores compared with patients receiving BSC alone. 34

Current UK guidelines and guidance

In terms of first-line treatment, NICE guidelines7 recommend that chemoradiation is the first choice of treatment for patients with stage IIIA disease and is also an option for patients with stage IIIB disease who have good PS (WHO 0–1 or KPS 80–100) and localised disease that can be safely encompassed in a radical radiotherapy treatment volume. These patients are a very different and much smaller group of patients than patients receiving palliative chemotherapy. Chemotherapy is the first choice of treatment for patients with stage IIIB and stage IV NSCLC assessed as being of WHO PS 0–1. 7 BSC (including palliative radiotherapy) is the first choice of treatment for patients with stage IV WHO PS 3–4, following medical optimisation.

The NICE guidelines7 recommend that first-line chemotherapy for advanced NSCLC should be a combination of a third-generation drug (DOC, GEM, PAX or VNB) and a platinum (PLAT) – either carboplatin (CARB) or cisplatin (CIS). According to the updated NICE guidelines,7 whether CARB or CIS is used depends on the balance of toxicity, efficacy and convenience. Patients who are unable to tolerate a PLAT combination may be offered single-agent chemotherapy with a third-generation drug. 7

Following STAs of PEM and gefitinib (Iressa^®, AstraZeneca; GEF), NICE also recommends that PEM plus CIS be considered as a first-line therapy for patients with locally advanced or metastatic NSCLC who are histologically confirmed as having large cell or adenocarcinoma. 13 GEF as a single agent is recommended as an option for the first-line treatment of patients with locally advanced or metastatic NSCLC who test positive for the EGFR-TK mutation. 16 Table 5 summarises the licensed indications and recommendations set out by NICE which govern the use of chemotherapy as a first-line treatment for patients with locally advanced or metastatic NSCLC in England and Wales.

The patient chemotherapy treatment pathway

Following the publication of guidelines by NICE in 2005,18 PLAT-based doublet chemotherapy has become established as the standard first-line treatment for patients with advanced NSCLC and good PS in the UK. Data from a large observational pan-European trial35 show that four cycles of PLAT-based chemotherapy treatment is standard practice in England and Wales. Figure 3 presents a flow diagram of the patient treatment pathway for first-line treatment of NSCLC and an estimate of the proportions of patients along the pathway. The proportion of patients who have non-squamous disease and are treated with PEM or GEM is unknown.

FIGURE 3.

Patient first-line chemotherapy treatment pathway in England and Wales. a, The Information Centre for Health and Social Care, 2007;12 b, Rossell et al. , 2009;17 c, best estimate; d, Epsicom Business Intelligence Ltd. 36,37 CTX, chemotherapy.

Availability of therapeutic agents

Table 6 lists the costs of available branded and generic preparations as taken from the British National Formulary (BNF). 38 AstraZeneca provides a Patient Access Scheme decreasing the cost of GEF to the NHS. In addition, clinical centres frequently negotiate prices below those listed in the BNF. 38 Over the past few years, the patents for a number of these agents have expired and they are now available in generic formulations which are less expensive. The currently available data on costs are discussed more fully in Chapter 4.

Identification of trials

The systematic review was guided by the general principles recommended in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. 40 A comprehensive search strategy was developed; search terms included a combination of index terms (e.g. non-small-cell lung carcinoma) and free-text words (e.g. lung cancer or lung tumour or lung carcinoma). The search of MEDLINE and EMBASE was restricted to papers with abstracts published in the English language. MEDLINE was searched from January 1990 to March week 3 2009 and EMBASE was searched from January 1990 to week 13 2009. The Cochrane Library (including Cochrane Reviews, Database of Abstracts of Reviews of Effects, Cochrane Central Register of Controlled Trials and Health Technology Assessments) was searched up to Issue 3, July 2010. An updated search was performed of MEDLINE and EMBASE to identify trials published up until August 2010. All references were exported to the EndNote version X4 (Thomson Reuters, CA, USA). Searches have been limited to these databases based on the evidence related to searching presented by Royle and Waugh,41 which demonstrates that wider searching is not always effective in retrieving additional trials for inclusion in a specific group of diseases including cancer. Details of the search strategies are available in Appendix 1.

The protocol was revised to exclude trials that had been published prior to the year 2000, owing to the large number of references identified by the searches and to reflect recent advances in chemotherapy treatments (e.g. third-generation chemotherapy drugs). The protocol is available in Appendix 2.

A number of hand searches were carried out to ensure the completeness of the review including the database of the American Society for Clinical Oncology (ASCO) and the US Food and Drug Administration (FDA) and EMA websites. A key review of chemotherapy treatments for patients with NSCLC by Clegg et al. 39 was searched for relevant trials. Reference lists of included trials were also searched to identify any further relevant trials.

Evidence synthesis

Analysis was restricted to chemotherapy drugs currently licensed in Europe and approved by NICE. Recent clinical evidence from PEM-based trials has suggested significant interaction between PEM efficacy and tumour histology and indicates that histology is critical in choosing the appropriate therapy for patients with NSCLC. Based on these data, NICE recommends that PEM in combination with CIS may also be considered as a first-line therapy for patients with locally advanced or metastatic NSCLC with disease histologically confirmed as large cell or adenocarcinoma. 13 Similarly, evidence from GEF-based trials indicates that GEF efficacy depends on the presence of sensitising EGFR mutations in the tumour. GEF is recommended by NICE as an option for the first-line treatment of locally advanced or metastatic NSCLC that tests positive for the EGFR-TK mutation. 16 Recent trial evidence, therefore, supports the concept that treatment choices in the first-line management of advanced NSCLC should no longer be the same for all patients with NSCLC, but rather decisions must take into consideration tumour histology subtyping and also molecular profiling.

To reflect current UK treatment pathways, analyses were undertaken and reported for three subpopulations of NSCLC: patients with predominantly squamous disease, patients with predominantly non-squamous disease and patients who were EGFR M+. In the main, all analyses were conducted on the total population according to randomisation; however, subpopulation data were included in our analyses if used previously for international or national decision-making.

Patients with squamous disease can be treated with any of the third-generation drugs (DOC, GEM, PAX or VNB) in combination with PLAT. Very few published RCTs differentiate between subpopulations of patients. We assume that the results of all studies that do not differentiate between subpopulations are equally applicable to patients with squamous disease and non-squamous disease. Before adopting this approach, we identified four third-generation studies43–46 that reported multivariate statistical testing and included histology as a candidate explanatory variable. From our critique of these studies, we concluded that there was no significant influence of histology on outcomes for patients with squamous or non-squamous disease. In this review, all data applicable to the squamous population were derived from mixed population studies; however, none of the studies included in the review investigated the use of chemotherapy solely for patients with squamous disease.

Patients with non-squamous disease who are not EGFR M+ can be treated with either third-generation drugs in combination with PLAT or PEM + CIS. This means that the data available to support treatment decisions for patients with non-squamous disease may be derived from analyses of total (mixed) population studies as well as from RCTs where survival analyses by histology may have been undertaken. Use of subpopulation data means that survival analyses were not conducted on the total trial population according to randomisation. Subpopulation data regarding the use of PEM have been used as the basis for the award of European marketing authorisation and regulatory decision-making, we therefore considered use of these subpopulation data in our analyses to be reasonable and appropriate.

Patients who are EGFR M+ can be treated with either third-generation drugs in combination with PLAT or GEF. Again, subpopulation data regarding the use of GEF have been used as the basis for the award of European marketing authorisation and regulatory decision-making, we therefore considered use of subpopulation data in our analyses to be reasonable and appropriate.

Data on any of the following outcomes were included in the meta-analyses: OS; survival at 1 and 2 years, PFS and TTP. Definitions of PFS and TTP varied between trials and the PFS and TTP outcomes in all of the included trials were assessed for eligibility for inclusion in analyses. No evidence synthesis was attempted for QoL, AEs and ORR owing to limited data or variability in outcome assessment.

Both direct head-to-head meta-analysis and mixed-treatment comparison approaches were undertaken in order to integrate information on the relative efficacy of all included drugs as an insufficient number of trials were available that directly compared all treatment options. When sufficient data permitted, analyses were undertaken for the squamous population, the non-squamous population and the EGFR M+ population.

For the analyses of the population with squamous disease, trials for PEM and GEF were excluded. Therefore, the following analyses on OS, PFS and TTP were planned. The primary analyses in the population with squamous disease were direct meta-analysis 1 and mixed-treatment comparison 1. The remaining analyses were all sensitivity analyses to explore the impact of particular trials or characteristics on results from the primary analyses: the sensitivity analyses were undertaken to explore the impact of six cycles of chemotherapy, different combinations of chemotherapy and PLAT, and trials with < 24 months follow-up.

Primary analyses: population with squamous disease:

• Direct meta-analysis 1: standard direct head-to-head meta-analysis using data from four licensed third-generation agents (PAX, VNB, DOC and GEM) from 18 trials. 47–60 This included four pair-wise meta-analyses for the following comparisons: GEM + PLAT compared with VNB + PLAT, GEM + PLAT compared with PAX + PLAT, VNB + PLAT compared with PAX + PLAT and VNB + PLAT compared with DOC + PLAT. Data for two comparisons (GEM + PLAT vs DOC + PLAT, PAX + PLAT vs DOC + PLAT) were available from single trials and, therefore, no direct meta-analysis was undertaken.

• Mixed-treatment comparison 1: mixed-treatment comparison using data from four licensed third-generation agents (PAX, VNB, DOC and GEM) from 18 trials. 47–60 The analysis included direct and indirect evidence from all six pair-wise comparisons: GEM + PLAT compared with VNB + PLAT, GEM + PLAT compared with PAX + PLAT, VNB + PLAT compared with PAX + PLAT, VNB + PLAT compared with DOC + PLAT, GEM + PLAT compared with DOC + PLAT and PAX + PLAT compared with DOC + PLAT.

Sensitivity analyses: population with squamous disease:

• Direct meta-analysis A and mixed-treatment comparison A: sensitivity analysis excluding Chen et al. 52 (< 24 months follow-up time) and Tan et al. 59 (used six cycles of chemotherapy).

• Direct meta-analysis B and mixed-treatment comparison B: sensitivity analysis using data from PAX + CIS instead of PAX + CARB from the Schiller et al. 47 trial.

• Direct meta-analysis C and mixed-treatment comparison C: sensitivity analysis using data from DOC + CARB instead of DOC + CIS from the Fossella et al. 44 trial.

• Direct meta-analysis D and mixed-treatment comparison D: sensitivity analysis excluding Tan et al. 59 (used six cycles of chemotherapy).

• Direct meta-analysis E and mixed-treatment comparison E: sensitivity analysis excluding Chen et al. 52 (< 24 months follow-up time). This affects VNB + PLAT compared with DOC + PLAT pair-wise comparison.

For the population with non-squamous disease, the following analyses on OS and PFS were planned.

Primary analyses: population with non-squamous disease:

• Direct meta-analysis 1: standard direct head-to-head meta-analysis using data from four licensed third-generation agents (PAX, VNB, DOC and GEM) from 18 trials47–60 and two PEM studies. 61,62 This included five pair-wise meta-analyses for the following comparisons: GEM + PLAT compared with VNB + PLAT, GEM + PLAT compared with PAX + PLAT, VNB + PLAT compared with PAX + PLAT, VNB + PLAT compared with DOC + LAT and GEM + PLAT compared with PEM + PLAT. Data for two comparisons (GEM + PLAT vs DOC + PLAT and PAX + PLAT vs DOC + PLAT) were available from single trials and, therefore, no direct meta-analysis was undertaken.

• Mixed-treatment comparison 1: mixed-treatment comparison using data from four licensed third-generation agents (PAX, VNB, DOC and GEM) from 18 trials47–60 and two PEM studies. 61,62 The analysis included direct and indirect evidence from all 10 pair-wise comparisons: GEM + PLAT compared with VNB + PLAT, GEM + PLAT compared with PAX + PLAT, GEM + PLAT compared with DOC + PLAT, GEM + PLAT compared with PEM + PLAT, VNB + PLAT compared with PAX + PLAT, VNB + PLAT compared with DOC + PLAT, VNB + PLAT compared with PEM + PLAT, PAX + LAT compared with DOC + PLAT, PAX + PLAT compared with PEM + PLAT and DOC + PLAT compared with PEM + PLAT.

The following sensitivity analyses were undertaken to explore the impact of six cycles of chemotherapy, different combinations of chemotherapy and PLAT, trials with < 24 months follow-up and the one study62 with PEM + CARB which is not licensed in the UK.

Sensitivity analyses: population with non-squamous disease:

• Direct meta-analysis A and mixed-treatment comparison A: sensitivity analysis excluding Chen et al. 52 (< 24 months follow-up time) and Tan et al. 59 (used six cycles of chemotherapy).

• Direct meta-analysis B and mixed-treatment comparison B: sensitivity analysis using data from PAX + CIS instead of PAX + CARB from the Schiller et al. 47 trial.

• Direct meta-analysis C and mixed-treatment comparison C: sensitivity analysis using data from DOC + CARB instead of DOC + CIS from the Fossella et al. 44 trial.

• Direct meta-analysis D and mixed-treatment comparison D: sensitivity analysis excluding Tan et al. 59 (used six cycles of chemotherapy).

• Direct meta-analysis E and mixed-treatment comparison E: sensitivity analysis excluding Chen et al. 52 (< 24 months follow-up time).

• Direct meta-analysis F and mixed-treatment comparison F: sensitivity analysis excluding Gronberg et al. 62 (contains PEM + CARB which is not licensed in the UK).

For the EGFR M+ population, the following analyses on OS and PFS were planned:

• Direct meta-analysis 1: standard direct head-to-head meta-analysis including two trials. 15,63,64

• Mixed-treatment comparison A: analysis including three GEF trials. 15,63–65

• This analysis includes both direct and indirect evidence for all three trials.

Direct evidence synthesis

All analyses for the NSCLC population with squamous disease were based on the intention-to-treat (ITT) population where possible and as noted above, included a mix of patients with squamous and non-squamous disease. For non-squamous and EGFR M+ populations, data from trials15,62,63 were based on the results of subgroup analyses. Where appropriate, standard meta-analysis were undertaken for each pair-wise treatment comparison using the ‘metan’ command within Stata Version 9.2 (StataCorp LP, College Station, TX, USA). For time-to-event outcomes (OS, PFS and TTP), the trial level estimate of log-hazard ratio (HR) and its variance were extracted directly from trial publications if available. Additional data were requested whenever needed from the authors of trials that directly compared first-line chemotherapy treatments currently licensed in Europe and approved by NICE. These additional data were requested in order to include as many relevant trials as possible in the meta-analysis. Details of additional data requested and provided are presented in Appendix 5. In the absence of direct estimates from published papers or requested from the authors, previously reported methods that use published data such as Kaplan–Meier survival curves or log-rank statistics were used to estimate the required trial-level log-HR and its variance. 66,67 A random-effects (frequentist) inverse variance-weighted approach was used to pool estimates of log-HR across trials.

In economic modelling, both short- and long-term survival data are always preferred when projecting survival benefits for a technology over a lifetime period using the best available evidence. This evidence is normally derived from a meta-analysis as the most appropriate summary statistic because this takes into account both the number of events and the time to these events, and also the data from those patients who have been censored. However, trial reports do not always report time to event data. Therefore, in order to address this, 1-year and 2-year survival data were extracted from trial reports, along with the number randomised to each treatment group to estimate the risk ratio within each trial and used a random effects Mantel–Haenszel approach to calculate the pooled risk ratio with a 95% confidence interval (CI). Although 1-year and 2-year survival analysis was specifically designed to inform economic modelling, this summary measure has some limitations, especially when follow-up and censoring patterns vary from trial to trial. One approach would be to adjust for variable follow-up and censoring across trials. However, in this analysis, not all trials reported the censoring rates and for simplicity, we assumed that these factors were comparable across trials.

Statistical heterogeneity was assessed by considering the chi-squared test for heterogeneity with a 10% level of significance, and the I²-statistic with a value of 50% representing moderate heterogeneity. 68,69

Mixed-treatment comparison – direct and indirect comparisons

As trials conducting head-to-head comparisons of all treatments under evaluation were not available or insufficient for some comparisons, the possibility of conducting an indirect comparison was investigated. This approach fulfils the objective of providing simultaneous comparison of all the relevant treatment alternatives, and can provide information about the associated decision uncertainty or sufficient information for economic evaluation. Hence, for the purposes of decision-making, a Bayesian mixed-treatment comparison framework was adopted to synthesise information on all technologies simultaneously using Markov Chain Monte Carlo (MCMC) methods to estimate the posterior distributions for our outcomes of interest. The MCMC simulation begins with an approximate distribution and, if the model is a good fit to the data, the distribution converges to the true distribution. The mixed-treatment comparison analysis allows for the synthesis of data from direct and indirect comparisons and allows for the ranking of different treatments in order of efficacy and estimation of the relative treatment effect of competing interventions. This approach assumes exchangeability of treatment effect across all included trials, such that the observed treatment effect for any comparison could have been expected to arise if it had been measured in all other included trials. This was assessed informally through examination of the trial populations and comparability of outcomes in the common treatment group facilitating the comparison. Inconsistency in the treatment effects between pair-wise comparisons were investigated by comparing the direct and indirect evidence together with the 95% CIs.

As with all meta-analyses, mixed-treatment comparison may be conducted using either fixed- or random-effects models. Random-effects models allow for the possibility that the true treatment effect may differ between trials. In our analyses, random-effects models were used throughout. Model fit was assessed based on residual deviance and deviance information criteria. Adjustment for multiarm trials was performed since estimates of relative treatment effects from trials with more than two treatment arms will be correlated owing to their joint dependence with the reference treatment arm.

In each MCMC simulation, we ranked the absolute log-hazard then used it to calculate the probability that each treatment was best across all simulations. 70,71 If a treatment is significantly better than all other treatments in the mixed-treatment comparison, the probability of it being the most effective treatment will be at least 95%. A probability < 95% indicates that there is at least one other treatment which is not significantly different to the best treatment (at the 5% level). A non-informative (flat prior) normal distribution was used for the log-HR and log-relative risk (RR) of each relative comparison; thus, the observed results are completely influenced by the data and not the choice of prior.

WinBUGS version 1.4 statistical software (MRC Biostatistics Unit, Cambridge, UK) was used for the mixed-treatment comparison analysis by adapting code (presented in Appendix 6) from the Multi-Parameter Evidence Synthesis Research Group. 72 Two chains were used to ensure that model convergence was met after 90,000 iterations with a burn-in of 10,000. Formal convergence of the models was assessed using trace plots and the Gelman–Rubin approach73 and through inspection of the history plots. OS, PFS and TTP results were expressed as HRs with 95% CI.

Quantity of research available

As shown in Figure 4, the electronic searches identified 5378 citations (Table 8 describes in detail the results of the database searching). Initial screening identified 330 potentially relevant references; these were obtained as full-text copies, and the 240 references that were published post 2000 were assessed for eligibility for inclusion. Of the 223 trials, 30 trials were excluded because they were not chemotherapy versus chemotherapy comparisons. Information regarding the 17 RCTs that were excluded from the 240 references found from electronic searching are listed, with reasons for exclusion, in Appendix 7.

FIGURE 4.

Flow diagram of inclusion of trials.

The database of abstracts from the ASCO annual NSCLC meetings up to and including 2010 was searched to identify any relevant trials from details of conference abstracts. Three abstracts were identified as potentially relevant for inclusion; however, full-text articles could not be found of any of the three abstracts. 74–76

Overall, 193 trials compared chemotherapy with chemotherapy, of which 23 trials (reported in 24 publications15,47–65) compared chemotherapy drugs currently licensed in Europe and recommended by NICE for the first-line treatment of patients with locally advanced or metastatic NSCLC.

Quality

Full details of the quality assessment criteria are presented in Appendix 4. Results of the methodological quality assessment of trials are presented in Table 9. Overall, methodological quality of included trials was poor. Only six15,43,45,55,57,64,65 of the 23 included trials reported sufficient information for them to be assessed as adequately randomised and with adequate concealment of random allocation.

All trials clearly reported the number of participants randomised. All trials reported eligibility criteria and, with the exception of four trials,48,56,58,60 all trials reported details about co-interventions, for example palliative radiotherapy and/or second-line chemotherapy (in one trial54 a minority of patients had surgery following chemotherapy). In the trial by Douillard et al. , 53 second-line therapy was built into the trial design; however, nearly all trials (appropriately) allowed second-line treatments which could potentially confound results. Five trials15,44,58,62,64,65 were reported as open, i.e. assessors, administrators and participants were not blinded; two studies59,63 blinded the assessors. In 16 studies43,45–57,60,61 the authors did not state whether or not blinding of participants, investigators or outcome assessors was carried out. The outcomes of > 80% of patients were assessed in all studies and all studies reported reasons for dropout; 10 trials15,44,48,49,51–53,59–61,64 used an ITT approach to assess OS. Five of the trials48,53,57,60,61 appeared to report fewer outcomes than initially stated.

Bias can occur as a result of the early closure of trials;77 it is noted that three of the included trials were stopped early. 47,49,65 In the trial by Gebbia et al. ,49 further accrual into the two ‘sequential chemotherapy’ arms [in this case meaning GEM + ifosfamide (Mitoxana^®, Baxter Healthcare Ltd) followed by VNB + CIS or the opposite sequence, which are two group comparisons not included within this review] was stopped because the VNB + CIS arm appeared to be more effective. Owing to initial slow accrual, the protocol of the trial by Mitsudomi et al. 65 was amended to include patients with stage IIIB/IV disease and to allow outsourcing of EGFR genetic testing in order to further facilitate patient accrual. Accrual was halted when investigators considered the trial to be sufficiently powered, making further accrual of patients unnecessary, and final analyses were done on the available data. In the trial by Schiller et al. ,47 after the first 68 patients, accrual of the PS 2 cohort was halted owing to a high incidence of AEs, including five deaths (subsequent analysis showed that only two of the five deaths were clearly treatment related). All trials provided reasons for withdrawal; see Table 23 for actual numbers of patients treated.

Trial characteristics

Trial characteristics are presented in Appendix 8. The 23 trials were published between 2001 and 2010. Of the 20 multicentre trials, six have international centres. 15,44–46,59,61,64 The three single-centre trials were all located in Taiwan. 50–52 All included trials were published in English.

There are five Phase II trials,51–53,56,58 16 Phase III trials15,43–46,48,49,54,55,57,59–65 and two trials47,50 with phase undefined. Ten trials15,43,44,53,57–62,64 were funded solely by pharmaceutical companies, five trials46,47,56,63,65 were funded by research grants, two trials45,48 were funded by both pharmaceutical companies and research grants and funding was not stated in six trials. 49–52,54,55

Seventeen trials15,43–48,51–54,57,60–62,64,65 were sufficiently powered to evaluate OS, four trials49,56,59,63 were inadequately powered and the power of two trials50,58 was unclear. (If a trial reported an estimated sample size and then randomised at least this number of patients, then the trial was assessed as sufficiently powered.) Median follow-up ranged from 11 to 40 months.

Details of trial interventions are presented in Appendix 9. Four trials43,44,46,60 compared three treatment arms and three trials47,49,57 compared four treatment arms (not all arms met the inclusion criteria for this analysis). Table 10 shows the chemotherapy comparisons which were available from the 23 included trials. Trials using either CARB or CIS are both described as including PLAT.

Doses of chemotherapy drugs used varied, the median number of chemotherapy cycles ranged from 2.6 to 6, and route of administration was intravenous (i.v.) or oral. The majority of trials reported second-line chemotherapy and/or palliative radiotherapy, with the exception of one trial47 that reported that second-line treatment data were not collected, and four trials49,56,58,65 in which it was unclear whether or not patients received any second-line treatment. Two trials48,54 reported that patients who went on to have radiotherapy were excluded from the analysis.

Patient characteristics

Patient characteristics are presented in Appendix 10. Trial patients are generally younger and have better PS and fewer comorbidities than patients in a clinical setting. Full details of individual trial inclusion criteria can be found in Appendix 11. The majority of patients were male with adenocarcinoma stage IIIB or IV and a PS of 1. Only five trials46,48,55,58,60 reported details of the staging system used to classify patients and given the variety in the dates and settings of the trials, the staging systems used is most likely to have varied across trials.

The number of patients randomised into trial arms ranged from 39 to 863. Median age ranged from 56 to 67 years within the clinical trials, which is younger than routinely found in clinical practice. The most common age group at diagnosis of ‘malignant neoplasm of bronchus or lung’ for men and women in England in 2008 was 75–79 years. 78 The percentage of males within each trial arm ranged from 56% to 84% for trials with PLAT-based doublets incorporating third-generation chemotherapy drugs. In the three GEF trials,15,63–65 the proportion of males to females is much less; the percentage of males ranged from 21% to 37%, this is because sex is a factor in the likelihood of the presence of EGFR mutation (mutation found more often in females).

The patient populations in the trials by Maemondo et al. 63 and Mitsudomi et al. 65 were quite different from those in the other trials. These two trials63,65 were based on molecular selection, and only patients with EGFR M+ tumours were eligible for inclusion. The Iressa Pan ASian Study (IPASS)15,64 restricted the patient population to those with adenocarcinoma who were never-smokers or former light smokers in order to increase the likelihood of the presence of the EGFR mutation. All three trials15,63–65 were multicentre, but conducted within East Asian countries. Patients assigned to the experimental arms received oral GEF at the standard dose (250 mg daily). Patients assigned to the chemotherapy arm received different PLAT-based doublets (PAX + CARB in two trials15,63,64 and DOC + CIS in one trial65).

The majority of patients within the trials had stage IIIB or IV disease; at least twice as many patients had stage IV disease as stage IIIB disease.

Outcomes

Population 1: non-small cell lung cancer patients with squamous disease

Eighteen trials43–60 were eligible for inclusion in the direct meta-analysis and mixed-treatment comparison analyses in the population with squamous disease with 7382 randomised patients. Comparisons between PLAT-based doublets incorporating third-generation chemotherapy drugs were available from 18 trials. 43–60 The characteristics of trials and patients included in these trials are described narratively in Assessment of clinical effectiveness. A summary of chemotherapy regimens showing treatment arms included in the evidence synthesis is presented in Table 14, and shows that the majority of trials (n = 16) had at least one treatment arm containing a CIS regimen; whereas, nine trials had at least one treatment arm containing a CARB regimen.

Seven trials43,44,47,48,56–58 directly compared CIS and CARB when used in combination with one of the third-generation chemotherapy drugs. The non-PLAT-based chemotherapy arms from four trials45,46,49,60 were excluded from all analyses because these treatments are not currently recommended by NICE in the first-line management of patients with advanced NSCLC in the UK.

Overall survival

Overall survival was defined consistently across trials as the time from randomisation to death from any cause. The data points included in the direct meta-analysis and mixed-treatment comparison analyses for OS are presented in Table 15 and six pair-wise comparisons are summarised in Figure 5. Analyses for OS were based on 18 trials43–60 involving 7382 randomly assigned patients and 6081 deaths. The data sources for OS HRs used in the direct meta-analysis and mixed-treatment comparison analyses are also displayed in Table 15. This indicates that not all trials reported HRs for OS. Thus, pre-specified methods (see Evidence synthesis) were used to extract the HR and its variance for each trial that reported any information on OS outcome. The HRs for OS from three trials43,44,57 were extracted directly from the trial papers, data from four trials51–53,58 were obtained by contacting the investigators. In addition, HRs for four trials45–47,50 were extracted from a systematic review79 of the literature as the HRs were not reported in the primary trial publication (three out of four investigators of the primary trials were also co-authors of this systematic review). The remaining seven trials48,49,54–56,59,60 did not report HRs for OS and could not be obtained from trial authors. A network of 18 connected RCTs with six different treatment comparisons is presented in Table 15 and Figure 5. The circles in Figure 5 represent different treatments, and the lines represent direct head-to-head trials informing each comparison. Unconnected circles indicate a lack of direct randomised comparison.

FIGURE 5.

Network of RCTs comparing chemotherapy in the treatment of first-line advanced NSCLC (OS) for the mixed-treatment comparison and meta-analysis in the population with squamous disease.

Result summaries for all pair-wise comparisons between interventions from the direct meta-analysis and the mixed-treatment comparison primary analyses are presented in Table 16. Individual trial results and overall pooled results from both analyses, where available, are also displayed as forest plots for each pair-wise comparison. Results for the mixed-treatment comparison sensitivity analyses of OS are presented in Appendix 12.

Gemcitabine plus platinum compared with vinorelbine plus platinum

Eight head-to-head RCTs43,45,49,50,54,55,57,58 were eligible for inclusion in the direct meta-analysis and mixed-treatment comparison for GEM + PLAT compared with VNB + PLAT, with 2152 randomised patients and 1702 deaths. Seven trials43,45,49,50,54,57,58 used a CIS-based regimen in at least one treatment arm. CARB-based regimens were used in two trials, in one trial55 comparing GEM + CARB with VNB + CARB and in another trial58 comparing GEM + CARB with VNB + CIS.

The HR and 95% CI for each trial are displayed in Figure 6 together with pooled results from meta-analysis 1 and mixed-treatment comparison analyses. Visual examination of Figure 6, the chi-squared test for heterogeneity (p = 0.972), and the I²-statistic (0%) all suggest very good consistency. The pooled OS estimate from meta-analysis 1 (HR = 1.08; 95% CI 0.98 to 1.20) shows a trend in favour of GEM + PLAT, and this was similar to, and consistent with, results of mixed-treatment comparison analyses (see Table 16 and Figure 6). However, as the CI includes a HR of unity we cannot exclude the possibility of no evidence of difference between GEM + PLAT and VNB + PLAT. The median OS of GEM + PLAT ranged from 6.4 to 14 months compared with 7.3 to 11.4 months for VNB + PLAT, with smaller trials showing larger median OS than bigger trials.

FIGURE 6.

Forest plot illustrating results of direct meta-analysis 1 and mixed-treatment comparison in terms of HRs and 95% CI of OS in trials comparing GEM + PLAT vs VNB + PLAT in the population with squamous disease. a, Same median OS reported for both treatment arms.

Gemcitabine plus platinum compared with paclitaxel plus platinum

Six head-to-head RCTs43,46,47,56,57,60 were eligible under this comparison, with up to 2589 patients and 2239 deaths. In this comparison, GEM was most frequently combined with CIS;43,46,47,56,57 only one trial60 used the GEM + CARB combination. Five trials43,47,56,57,60 evaluated the efficacy of PAX + CARB and two trials46,47 evaluated PAX + CIS. One multiarm trial47 evaluated efficacy for both PAX + CIS and PAX + CARB; however, in our analyses we excluded the PAX + CIS arm from this trial because of limited data points and included the PAX + CARB arm for both direct meta-analysis and mixed-treatment comparison analyses. A sensitivity analysis using PAX + CIS data produced similar results to the direct meta-analysis and the mixed-treatment comparison analyses (see Appendix 12). The HR and 95% CI for each trial are displayed in Figure 7 together with pooled results from meta-analysis 1 and mixed-treatment comparison analyses. Visual examination of Figure 7, the chi-squared test for heterogeneity (p = 0.948), and the I²-statistic (0%) all suggest very good consistency. The pooled OS estimate from meta-analysis 1 (HR = 1.03; 95% CI 0.94 to 1.13) was not statistically significant, and this was similar to and consistent with results from meta-analysis 2 and mixed-treatment comparison analyses (see Table 16 and Figure 7). The results provide insufficient evidence to support a difference in OS between GEM + PLAT and PAX + PLAT treatment. The median OS of GEM + PLAT-treated patients ranged from 6.2 to 14 months compared with 6.9 to 12.3 months in PAX + PLAT trials.

FIGURE 7.

Forest plot illustrating results of direct meta-analysis 1 and mixed-treatment comparison in terms of HRs and 95% CI of OS in trials comparing GEM + PLAT vs PAX + PLAT in the population with squamous disease.

Vinorelbine plus platinum compared with paclitaxel plus platinum

Four head-to-head RCTs43,48,51,57 were eligible for inclusion in this comparison, with 1228 patients and 977 deaths. CIS was used in combination with VNB in all trials; thus, the pooled results from this comparison can be treated as the efficacy of VNB + CIS. Three trials43,48,57 evaluated the efficacy of PAX + CARB; only one trial51 evaluated PAX + CIS.

The HR and 95% CI for individual trials are displayed in Figure 8 together with the pooled results from meta-analysis 1 and mixed-treatment comparison analyses. Visual examination of Figure 8, the chi-squared test for heterogeneity (p = 0.802), and the I²-statistic (0%) all suggest very good consistency. The pooled OS estimate from meta-analysis 1 (HR = 0.98; 95% CI 0.83 to 1.16) was not statistically significant, and was similar and consistent with results from mixed-treatment comparison analyses (see Figure 8). Overall, these results provide insufficient evidence to support a difference in OS between VNB + PLAT and PAX + PLAT. Median OS associated with PAX + PLAT ranged from 8.6 to 12.3 months compared with 8.1 to 15.4 months in the VNB + PLAT trials (see Figure 8).

FIGURE 8.

Forest plot illustrating results of direct meta-analysis 1 and mixed-treatment comparison in terms of HRs and 95% CI of OS in trials comparing VNB + PLAT vs PAX + PLAT in the population with squamous disease.

Vinorelbine plus platinum compared with docetaxel plus platinum

Four head-to-head RCTs44,52,53,59 were eligible for inclusion in this comparison, with 1525 patients and 1941 deaths. The PLAT regimen used in the four trials was CIS combined with VNB or DOC; thus, the pooled results from this comparison can be treated as the efficacy of VNB + CIS compared with DOC + CIS. One of these trials44 also evaluated the efficacy of DOC + CARB in addition to two CIS-based arms; this arm was excluded from the analysis because of insufficient trial data on the DOC + CARB treatment combination. However, we tested the use of the DOC + CARB arm in a sensitivity analysis and the results of the analysis did not show any significant difference.

The HR and 95% CI for individual trials are displayed in the in Figure 9 together with the pooled results from meta-analysis 1 and mixed-treatment comparison analyses. Visual examination of Figure 9, the chi-squared test for heterogeneity (p = 0.906), and the I²-statistic (0%) all suggest very good consistency. The pooled OS estimate from meta-analysis 1 (HR = 0.89; 95% CI 0.78 to 1.00) suggest an advantage to DOC + PLAT with similar and consistent results from mixed-treatment comparison analyses (see Table 16 and Figure 8). However, the CI includes values of HR that may not be clinically important. Median OS for DOC + PLAT ranged from 8.3 to 13 months compared with 9 to 13.8 months in the VNB + PLAT trials (see Figure 9).

FIGURE 9.

Forest plot illustrating results of direct meta-analysis 1 and mixed-treatment comparison in terms of HRs and 95% CI of OS in trials comparing VNB + PLAT vs DOC + PLAT in the population with squamous disease.

Progression-free survival

The PFS and TTP outcomes in all of the 18 trials considered to be eligible for inclusion in the analysis of the population with squamous disease were reviewed. Eleven trials43,44,46,48–50,55,57–60 were excluded from the PFS analysis since non-standard definitions of PFS appeared to be utilised by the investigators. Seven trials45,47,51–54,56 were included in the PFS analysis detailed in Table 17; however, most of these trials used slightly different definitions of PFS. For instance, in six trials45,47,51–54 PFS was referred to as TTP despite inclusion of death (as in the standard definition of PFS). This was particularly evident in trials designed before PFS was officially recognised as an appropriate surrogate for OS by the US FDA in 2007. 80

A network of seven connected RCTs45,47,51–54,56 with six different treatment comparisons is presented in Table 18 and Figure 10. The circles in Figure 10 represent different treatments, and the lines represent direct head-to-head trials informing each comparison. Unconnected circles indicate a lack of direct randomised comparison.

FIGURE 10.

Network of RCTs comparing chemotherapy in the treatment of first-line advanced NSCLC PFS for the meta-analysis and mixed-treatment comparison analyses of the population with squamous disease. Total number of trials adds to nine because one trial47 had more than one treatment arm.

The data points included in the direct meta-analyses and mixed-treatment comparison analyses for PFS are presented in Table 18 and six pair-wise comparisons are summarised in Figure 10. Analysis of PFS was based on seven trials involving 3523 patients. 45,47,51–54,56 The HRs for PFS for three trials51–53 were obtained by contacting the investigators. Data from two trials45,47 were extracted from a systematic review79 as HRs were not reported in the primary trial publication (as noted earlier, three out of four investigators in the primary trials were also co-authors for this systematic review). The remaining two trials54,56 did not report HRs for PFS and so the HRs and associated variance were extracted using information on PFS outcome by applying pre-specified methods66,67 as described in Evidence synthesis.

Table 19 shows pair-wise comparison results related to GEM + PLAT, VNB + PLAT, PAX + PLAT and DOC + PLAT from the direct meta-analyses and the mixed-treatment comparison analyses. Where appropriate, direct estimates for each included trial and overall pooled HRs from both sets of analyses are also displayed as forest plots within each pair-wise comparison section. Results for the mixed-treatment comparison sensitivity analyses of PFS are presented in Appendix 13.

Gemcitabine plus platinum compared with vinorelbine plus platinum

Two head-to-head RCTs45,54 were eligible for this comparison, with 544 randomised patients. Both trials used CIS as the PLAT-based regimen; thus, this analysis can be considered as a comparison of GEM + CIS with VNB + CIS.

The HR and 95% CI for each trial are displayed in Figure 11 together with the pooled results from meta-analysis 1 and mixed-treatment comparison analyses. Visual examination of Figure 11, the chi-squared test for heterogeneity (p = 0.943), and the I²-statistic (0%) all suggest very good consistency. The pooled PFS estimate from meta-analysis 1 (HR = 1.09; 95% CI 0.87 to 1.38) was not statistically significant, and was similar and consistent with results from mixed-treatment comparison analyses (see Table 19 and Figure 11). These findings suggest lack of evidence to support any difference in PFS between GEM + CIS and VNB + CIS. Median PFS estimates were 5 months in both arms in one trial54 and appear to be similar across the two trials.

FIGURE 11.

Forest plot illustrating results of meta-analysis 1 and mixed-treatment comparisons in terms of HRs and 95% CI for PFS in trials comparing GEM and VNB in combination with CIS in the population with squamous disease. a, Median PFS reported for both arms (5.3 months).

Gemcitabine plus platinum compared with paclitaxel plus platinum

Two head-to-head RCTs47,56 including 1001 patients contributed to this analysis. Both trials used CIS as the PLAT agent; thus, this analysis can be considered as a comparison of GEM + CIS with PAX + CIS. The HR and 95% CI for each trial are displayed in Figure 12 together with the pooled result from meta-analysis 1 and mixed-treatment comparison analyses. Visual examination of Figure 12, the non-significant chi-squared test for heterogeneity (p = 0.670), and the I²-statistic (0%) all suggest very good consistency. The pooled direct HR (HR = 1.17; 95% CI 1.00 to 1.36; meta-analysis 1) suggests improvement in PFS for GEM + CIS over PAX + CIS, with a borderline statistically significant difference. The direct evidence is consistent with the results of the mixed-treatment comparison analyses. In addition, median PFS estimates were similar in both trials ranging from 3.0 to 4.2 months.

FIGURE 12.

Forest plot illustrating results of direct meta-analysis and mixed-treatment comparison in terms of HRs and 95% CI of PFS in trials comparing GEM and PAX in combination with CIS in the population with squamous disease.

Vinorelbine plus platinum compared with docetaxel plus platinum

Two trials52,53 involving 333 patients explored the role of VNB + PLAT compared with DOC + PLAT. Both trials used CIS; thus, this analysis can be considered as a comparison of VNB + CIS with DOC + CIS. The HR and 95% CI from each trial are displayed in Figure 13 together with the pooled HR estimates from meta-analysis 1 and the mixed-treatment comparison analyses. Visual examination of Figure 13, the non-significant chi-squared test for heterogeneity (p = 445) and the I²-statistic (0%) all suggest very good consistency. PFS appeared favourable to DOC + CIS over VNB + CIS, although this was not statistically significant (HR = 0.92; 95% CI 0.74 to 1.16). Similar findings were observed in the mixed-treatment comparison analyses (see Table 19 and Figure 13). These results indicate insufficient evidence to conclude whether or not there are differences in PFS between VNB + CIS and DOC + CIS. Median PFS estimates in both arms were similar and ranged from 5 to 6.3 months in the VNB arms and 4.7 to 5 months in the DOC arms.

FIGURE 13.

Forest plot illustrating results of direct meta-analysis and mixed-treatment comparison in terms of HRs and 95% CI of PFS in trials comparing DOC and VNB in combination with PLAT in the population with squamous disease.

Time to disease progression

Time to disease progression was reported in 7 out of 21 trials. 43,44,46,49,50,58,60 Time to disease progression in RCTs is usually defined as the time from randomisation until objective tumour progression, and does not include death from other causes. We allowed these seven trials to be analysed as a group as their definitions of TTP were similar (Table 20).

The data points included in the direct meta-analysis and mixed-treatment comparison analyses for TTP are presented in Table 21. Analyses of TTP were based on outcomes for 3572 randomised patients and approximately 1258 progressive events. Data on log-HR and its variance from two trials43,44 were extracted directly from the published trials, data from two trials were obtained by contacting investigators50,58 and data from one trial46 were extracted from a previously published meta-analysis. 79 Two out of seven trials did not report HRs for TTP49,60 and these were estimated via the pre-specified approaches66,67 described in Evidence synthesis.

The network of comparisons for TTP showing seven connected RCTs43,44,46,49,50,58,60 with six different treatment comparisons are presented in Table 22 and Figure 14. The nodes in Figure 14 represent different treatments and the lines represent direct head-to-head trials informing each comparison. Unconnected nodes indicate lack of direct evidence. The data overview indicate that PLAT-based doublets incorporating GEM, VNB, PAX and DOC generally had at least one data point and had been compared against each other.

FIGURE 14.

Network of RCTs comparing chemotherapy in the treatment of first-line advanced NSCLC TTP for the meta-analysis and mixed-treatment comparison analyses of the population with squamous disease. Total number of trials adds to nine because one trial43 had more than one treatment comparator.

Table 23 shows pair-wise comparison results between GEM + PLAT, VNB + PLAT, PAX + PLAT and DOC + PLAT from the direct meta-analyses and the mixed-treatment comparison analyses. Direct estimates for each included trial and overall pooled HR from both analyses are also displayed as forest plots within each pair-wise comparison section. Results for the mixed-treatment comparison sensitivity analyses of TTP are presented in Appendix 14. Appendix 15 shows results for the mixed-treatment comparison sensitivity analyses of PFS and TTP combined.

Gemcitabine plus platinum compared with vinorelbine plus platinum

Four head-to-head RCTs43,49,50,58 were eligible under this comparison with 869 randomised patients. The PLAT-based regimen for VNB combination in all trials was CIS. For the GEM combination, one trial58 had CARB as the PLAT-based regimen. The HR and 95% CI for each trial are displayed in Figure 15 together with the pooled results from meta-analysis 1 and mixed-treatment comparison analyses. Visual examination of Figure 15, the chi-squared test for heterogeneity (p = 0.992) and the I²-statistic (0%) all suggest very good consistency. The pooled TTP estimates from meta-analysis 1 (HR = 1.03; 95% CI 0.90 to 1.18) were not statistically significant, and this was similar and consistent with results from mixed-treatment comparison analyses (see Table 23 and Figure 15). The estimated point estimate for the TTP HR was slightly different but also consistent with results from meta-analysis 1 and mixed-treatment comparison analyses. These results suggest that there is no evidence to support any difference in TTP between GEM + PLAT and VNB + PLAT. Median TTP estimates ranged from 4.1 to 5.3 months and 4 to 6.6 months in the VNB + PLAT and GEM + PLAT arms, respectively (see Figure 15).

FIGURE 15.

Forest plot illustrating results of direct meta-analysis 1 and mixed-treatment comparison 1 in terms of HRs and 95% CI of TTP in trials comparing GEM and VNB in combination with PLAT in the population with squamous disease.

Gemcitabine plus platinum compared with paclitaxel plus platinum

Three head-to-head RCTs PLAT43,46,60 including 1486 patients contributed to this analysis. The treatment arm PLAT combinations used in the trials were not very common among the included trials. For instance, GEM + CIS was used in two trials,43,46 GEM + CARB in one trial,60 PAX + CIS in one trial46 and PAX + CARB in two trials. 43,60 The HR and 95% CI for each trial are displayed in Figure 16 together with the pooled meta-analysis 1 result and HRs from the mixed-treatment comparison analyses. Visual examination of Figure 16, the non-significant chi-squared test for heterogeneity (p = 0.691) and the I²-statistic (0%) all suggest very good consistency. The pooled TTP estimate from the meta-analysis 1 (HR = 1.01; 95% CI 0.90 to 1.13) was not statistically significant, and this was slightly different but consistent with result from mixed-treatment comparison 1 analysis (see Table 19). These results suggest there is no evidence to support any difference in TTP between GEM + PLAT and PAX + PLAT. Median TTP estimates ranged from 4.3 to 5.3 months and 4.2 to 5.5 months in the GEM + PLAT and PAX + PLAT arms, respectively (see Table 19 and Figure 16).

FIGURE 16.

Forest plot illustrating results of direct meta-analysis 1 and mixed-treatment comparison 1 in terms of HRs and 95% CI of TTP in trials comparing GEM and PAX in combination with PLAT.

Survival risk at year 1 and year 2 post randomization

Year 1 and year 2 survival risk were defined as the probability of survival in intervals of time elapsed from randomisation to year 1 and year 2, respectively. Analyses were based on 17 trials43,44,46–60 involving 7136 randomly assigned patients. There was insufficient information on survival risk at 1 or 2 years for one trial. 45 The proportion of patients still alive or survival rates at year 1 or 2 for all trials were extracted directly from the published reports or indirectly from the survival curves in the published reports.

Results for meta-analysis and mixed-treatment comparison analyses for 1-year and 2-year survival are shown in Appendices 16 and 17, respectively. Analyses for 1-year survival show no evidence to suggest any difference in survival between the third-generation chemotherapy treatments. None of the results from the 2-year analyses were statistically significant with wide CIs that include clinically important values.

Population 2: non-small cell lung cancer patients with non-squamous disease

Current treatment pathways in the UK show that patients with non-squamous locally advanced or metastatic NSCLC can receive any of the four PLAT-based third-generation chemotherapies (GEM, VNB, PAX or DOC) as a first-line treatment. At the time of the initial guideline recommendations,18 the clinical effectiveness of these chemotherapy regimens for non-squamous histology was unknown as clinical effectiveness data were not assessed according to histology. In addition, patients with non-squamous disease can receive PEM + CIS. Despite recommendations for use of these treatments, no comprehensive trial has directly compared all of these treatments in this patient population. Moreover, there are currently only two trials that have direct evidence comparing PEM + CIS61 or PEM + CARB62 with any of the four PLAT-based combinations (i.e. GEM + CIS61 and GEM + CARB62). In this section, comparisons are attempted between the five drugs currently available to patients with non-squamous disease. It is assumed that the treatment effect for all PLAT-based third-generation chemotherapies is not dependent on histology as in the current NICE guideline. 7 The same data points for PLAT-based third-generation chemotherapies that were used for NSCLC patients with squamous disease (n = 18) with the addition of subgroup data from two PEM + PLAT trials were employed. 61,62 Analysis of TTP alone was not performed since none of the PEM studies presented data on TTP.

Table 24 presents a network of trials showing direct evidence on at least one of the outcomes of interest for all chemotherapy trials in the population with non-squamous disease.

Overall survival

The data points included in the direct meta-analysis and mixed-treatment comparison analyses for OS are described in the previous section on squamous population analyses (see Overall survival) in Table 15 and the 15 pair-wise comparisons are summarised in the network diagram (see Figure 5). Analyses for OS were based on all 20 trials, involving 9553 randomly assigned patients and 7608 deaths. OS data used in the meta-analyses and mixed-treatment comparisons for the analyses in the non-squamous disease population were derived from the same 18 trials as the data used in the analyses in the squamous disease population, except for PEM, in which case the data used reflect its licensed population (patients with adenocarcinoma and large cell). Table 25 presents a network of trials showing direct OS evidence for all chemotherapy trials in the population with non-squamous disease. Figure 17 presents a network of trials with OS data used in the mixed-treatment comparison and meta-analysis in the population with non-squamous disease. Result summaries for all pair-wise comparisons between interventions from the direct meta-analysis and the mixed-treatment comparison analyses are presented in Table 26. Individual trial results and overall pooled results from both analyses are displayed as forest plots for each pair-wise comparison where possible. Results from the mixed-treatment comparison sensitivity analyses for OS are shown in Appendix 18.

FIGURE 17.

Network of RCTs comparing chemotherapy in the treatment of first-line advanced NSCLC (OS) for the mixed-treatment comparison and meta-analysis in the population with non-squamous disease.

Results of direct meta-analysis and mixed-treatment comparison for OS in trials in the population with non-squamous disease are shown in Table 26. GEM + PLAT compared with VNB + PLAT had the greatest number of head-to-head trials (eight trials43,45,49,50,54,55,57,58). There was no direct head-to-head trial in population with non-squamous disease for three comparisons; thus, relative treatment effects for these three comparisons are derived entirely from the indirect estimates of the mixed-treatment comparison analysis.

Gemcitabine plus platinum compared with pemetrexed plus platinum

Two head-to-head RCTs61,62 with 2171 patients were eligible for comparison between GEM and PEM in combination with PLAT in the analysis of the population with non-squamous disease. The two PEM trials61,62 used different PLAT-based regimens, with one trial61 comparing PEM + CIS with GEM + CIS, and the other trial62 comparing PEM + CARB with GEM + CARB.

The HR and 95% CI for each trial are displayed in Figure 18 together with the pooled results from direct meta-analysis 1 and mixed-treatment comparison analyses. Visual examination of Figure 18, the chi-squared test for heterogeneity (p = 0.253) and the I²-statistic (23.4%) all suggest good consistency; the test for heterogeneity is non-significant and the I²-statistic is well below our pre-defined 50% cut-off point for moderate heterogeneity. The pooled OS estimate from meta-analysis 1 (HR = 0.85; 95% CI 0.73 to 1.00) was borderline statistically significant. The mixed-treatment comparison 1 result was statistically significant (HR = 0.85; 95% CI 0.74 to 0.98). These results suggest that there is evidence to support a difference in an OS benefit PEM + PLAT compared with GEM + PLAT. Median OS estimates were quite different across the two trials: in the Scagliotti et al. trial61 median OS was 11.8 compared with 10.4 months and in the Gronberg et al. trial62 median OS was 7.8 compared with 7.5 months for GEM + PLAT compared with PEM + PLAT, respectively. This difference could be because the proportion of patients who received postprogression treatment in the Scagliotti et al. trial61 was much higher than in the Gronberg et al. trial62 (54% and 32%, respectively). Alternatively, it could also be owing to other differences between the trials that we have not been able to explore.

FIGURE 18.

Forest plot illustrating results of direct meta-analysis and mixed-treatment comparison in terms of HRs and 95% CI of OS in trials comparing GEM + PLAT vs PEM + PLAT in the population with non-squamous disease.

Progression-free survival

The same data points used in the PFS analysis for the NSCLC population with squamous disease were used in this analysis except that the non-squamous specific PFS estimates for PEM + PLAT compared with GEM + PLAT from the two PEM trials were used. 61,62 Table 27 shows pair-wise comparison results between GEM + PLAT, VNB + PLAT, PAX + PLAT, DOC + PLAT and PEM + PLAT from the direct meta-analyses and the mixed-treatment comparison analyses. Eight trials45,47,51–54,56,61 were included in the PFS analysis detailed in Table 28. Results from the mixed-treatment comparison sensitivity analyses for OS are shown in Appendix 19. Appendices 20 and 21 shows results for direct meta-analysis and mixed-treatment comparison for combined PFS/TTP analyses.

The data points included in the direct meta-analyses and mixed-treatment comparison analyses for PFS are presented in Table 18 and 10 pair-wise comparisons are summarised in Figure 19. Analysis of PFS was based on eight trials involving 4396 patients. The HRs for PFS from one trial61 were extracted from the trial papers and data from three trials51–53 were obtained by contacting the investigators. Data from two trials45,47 were extracted from a systematic review,79 as HRs were not reported in the primary trial publication (as noted earlier, three out of four investigators in the primary trials were also co-authors for this systematic review). The remaining two trials did not report HRs for PFS54,56 and so the HRs and associated variance were extracted using information on PFS outcome by applying pre-specified methods66,67 as described in Evidence synthesis.

FIGURE 19.

Network of RCTs comparing chemotherapy in the treatment of first-line advanced NSCLC PFS for the meta-analysis and mixed-treatment comparison analyses of the population with non-squamous disease.

A network of eight connected RCTs45,47,51–54,56,61 with 10 different treatment comparisons are presented in Table 29 and Figure 19. The circles in Figure 19 represent different treatments, and the lines represent direct head-to-head trials informing each comparison. Unconnected circles indicate a lack of direct randomised comparison. There was no direct evidence for the following comparisons: VNB + PLAT compared with PEM + PLAT; PAX + PLAT compared with PEM + PLAT; and DOC + PLAT compared with PEM + PLAT. Relative treatment effects for these comparisons are estimated entirely from the indirect evidence available from the mixed-treatment comparison analysis.

Survival risk at year 1 and year 2 post randomization

Year 1 and year 2 survival risks were defined as the probability of survival in intervals of time elapsed from randomisation to years 1 and 2, respectively. The same data points used in the year 1 and year 2 analyses for the NSCLC population with squamous disease were used in this analysis except that the non-squamous specific estimates for PEM + PLAT compared with GEM + PLAT from the two PEM trials were used. 61,62 Results for meta-analysis and mixed-treatment comparison analyses for 1-year and 2-year survival are shown in Appendices 22 and 23, respectively. None of the results from the 2-year analyses were statistically significant with wide CIs that include clinically important values.

Epidermal growth factor receptor mutation-positive population

Three RCTs15,63–65 that compared GEF to PLAT-based chemotherapy as a first-line treatment of patients with advanced NSCLC were eligible for inclusion in the analysis of the EGFR M+ population. This is the first time that data from the Mitsudomi et al. 65 and Maemondo et al. 63 trials have been used in any meta-analysis or mixed-treatment comparison analyses within this report. All three trials15,63–65 were conducted in patients from East Asian populations who were identified as having adenocarcinoma, and being never or light smokers. Patients were randomised to GEF arms or to chemotherapy. Those in chemotherapy arms received different PLAT-based combinations: PAX + CARB in two trials15,63,64 and DOC + CIS in one trial. 65 Patient characteristics in the three trials15,63–65 appear to be similar; however, the selection of patients differed across the three trials. In the IPASS15,64 patient enrolment was not restricted to patients who were EGFR M+, whereas in the two trials63,65 only EGFR M+ patients were randomised. Therefore, the EGFR M+ data used in this report from the IPASS15,64 are restricted to the subgroup of patients classified as EGFR M+. OS and PFS were reported in all three trials; however, TTP and survival rates were not reported by EGFR M+ status.

Overall survival

Overall survival was defined consistently across the three trials15,63–65 as time from randomisation to death from any cause. The data points included in the direct meta-analyses and mixed-treatment comparison analyses for OS are presented in Table 30 and three pair-wise comparisons are summarised in Figure 20. Analysis for OS was based on the results of all three trials, involving 663 randomly assigned patients. HRs for OS for two trials15,64,65 were extracted from the trial papers and OS data for one trial63 were obtained by contacting the trial investigator. Three treatments (GEF, PAX + PLAT and DOC + PLAT) qualified for inclusion in the analysis of the EGFR M+ population. A network of connected RCTs with different treatment comparisons is presented in Table 30 and Figure 20. The nodes in Figure 20 represent different treatments, and the lines represent direct head-to-head trials informing each comparison. Unconnected nodes indicate lack of direct randomised comparison. There were two direct head-to-head comparisons PAX + PLAT compared with GEF15,63,64 and DOC + PLAT compared with GEF. 65 There was no direct evidence for the PAX + PLAT compared with DOC + PLAT comparison.

FIGURE 20.

Network of RCTs comparing GEF and chemotherapy used in the meta-analysis and mixed-treatment comparison analyses in the EGFR M+ population.

Result summaries for all pair-wise comparisons between interventions from the direct meta-analyses and the mixed-treatment comparison analyses including individual trial results are presented in Table 31.

Paclitaxel plus platinum compared with gefitinib

Two head-to-head RCTs15,63,64 including 484 patients were available that compared GEF and PAX + PLAT and contributed to the OS analysis in the EGFR M+ population. Both trials used CARB as the PLAT agent; thus, this analysis can be considered to compare PAX + CARB with GEF. The HR and 95% CI for each trial are displayed in Table 31 and Figure 21 together with the pooled meta-analysis result and HRs from the mixed-treatment comparison analyses. Visual examination of Figure 21, the non-significant chi-squared test for heterogeneity (p = 0.394) and the I2-statistic (0%) all suggest very good consistency. The pooled direct HR (HR = 0.94; 95% CI 0.74 to 1.18; meta-analysis for PAX + PLAT vs GEF) shows no significant difference in OS between GEF and PAX + PLAT. The direct meta-analysis evidence is consistent with the results of the mixed-treatment comparison analyses.

FIGURE 21.

Forest plot illustrating results of direct meta-analysis and mixed-treatment comparison in terms of HRs and 95% CI of OS in trials comparing GEF vs PAX + PLAT and GEF vs DOC + PLAT in the EGFR M+ population.

Progression-free survival

The definition of PFS was consistent across the three trials15,63–65 and all trials adopted Response Evaluation Criteria in Solid Tumours (RECIST)81 to assess tumour progression. The HRs for PFS were available from all trials. The data points included in the direct meta-analyses and mixed-treatment comparison analyses for PFS are presented in Table 32 and three pair-wise comparisons are summarised in Figure 22. Analysis for PFS was based on outcome data for 660 randomly assigned patients. Three treatments (GEF, PAX + PLAT and DOC + PLAT) qualified for inclusion in the analysis in the EGFR M+ population. A network of connected RCTs with different treatment comparisons is presented in Table 32 and Figure 22. The nodes represent different treatments, and the lines represent direct head-to-head trials informing each comparison. Unconnected nodes indicate lack of direct randomised comparison. There were two direct head-to-head comparisons (PAX + PLAT vs GEF and DOC + PLAT vs GEF). There was no direct evidence for PAX + PLAT compare with DOC + PLAT.

FIGURE 22.

Network of RCTs comparing GEF and chemotherapy used in the PFS meta-analysis and mixed-treatment comparison analyses in the EGFR M+ population.

Result summaries for all pair-wise comparisons between interventions from the direct meta-analyses and the mixed-treatment comparison analyses including individual trials results are presented in Table 33 and Figure 23.

FIGURE 23.

Forest plot illustrating results of direct meta-analyses and mixed-treatment comparisons in terms of HRs and 95% CI of PFS in trials comparing GEF vs PAX + PLAT and GEF vs DOC + PLAT in the EGFR M+ population.

Adverse events

This review presents data on AEs that were categorised in the published trials as being grade 3 and 4. Appendix 24–26 provides details of the proportion of patients who experience grade 3–4 AEs within each individual trial and toxic deaths reported within each trial.

The trials reported a diverse range of AEs and the definitions of AEs (including grading) varied between trials, making it difficult to summarise AE data. Tables 34–37 show statistically significant AEs reported within the trials by pair-wise group comparisons.

Table 34 shows that five trials43,50,54,55,58 report significantly higher levels of thrombocytopenia in GEM + PLAT arms compared with VNB + PLAT arms. However, four trials43,50,54,58 report significantly greater levels of neutropenia in VNB + PLAT arms compared with GEM + PLAT arms.

Table 35 indicates that haematological toxicity is more common in patients receiving GEM + PLAT (anaemia, blood transfusions, haemorrhage and thrombocytopenia) compared with patients receiving PAX + PLAT.

Table 36 indicates that haematological toxicity is more common in patients treated with VNB + PLAT compared with patients treated with PAX + PLAT.

Table 37 shows that anaemia and febrile neutropenia were significantly more common in patients treated with VNB + PLAT compared with patients treated with DOC + PLAT. Diarrhoea and alopecia are more common in patients treated with DOC + PLAT compared with patients treated with VNB + PLAT.

Other data relating to AEs, including details of treatment administration and relative dose intensity (RDI), are presented in Appendix 27. Trials reported details of median time to complete treatment, percentage of patients who completed treatment as per protocol, details of chemotherapy dose reductions and delays, and median number of chemotherapy cycles.

The number of patients discontinued who treatment because of toxicity was significantly higher in the VNB + CIS treatment arm than in the PAX + CARB or DOC + CIS arms. In the trial by Kelly et al. ,48 discontinuation was significantly higher, and completion of treatment and RDI significantly lower, in the VNB + CIS arm than in the PAX + CARB arm. In the trial by Fossella et al. ,44 patients in the DOC + CIS and the DOC + CARB arms had a higher median number of chemotherapy cycles, higher RDI and completion rates and fewer treatment delays than those in the VNB + CIS arm. Patients in the DOC + CIS arm of the trial by Douillard et al. 53 had a higher median RDI, fewer cycle delays and fewer chemotherapy dose reductions compared with the VNB + CIS arm.

There was higher RDI for GEM compared with VNB in the trial by Thomas et al. 58 However, in a trial by Helbekkmo et al. ,55 a significantly greater percentage of patients in the GEM arm had > 24 days between chemotherapy courses and delayed or cancelled chemotherapy at day 8 due to haematological toxicity compared with the VNB arm.

In the trial by Scagliotti et al. ,61 dose adjustments were less frequent and RDI was higher in the PEM arm than in the GEM arm. In the trial by Gronberg et al. ,62 the mean number of cycles was higher and significantly more patients in the PEM arm than in the GEM arm completed four cycles, and without delays.

In the trial by Schiller et al. ,47 treatment with GEM + CIS was more likely to cause grade 3, 4 or 5 renal toxicity and 27% of patients who received GEM + CIS were withdrawn from the trial owing to complications of therapy, compared with 15% of patients in the PAX + CIS arm (p< 0.001).

Gefitinib is associated with significantly lower severe toxic AEs compared with PAX + CARB15,63,64 and DOC + CIS65 with the exception of liver dysfunction. 65 In one trial, GEF15,64 was associated with a lower rate of AEs leading to discontinuation of the drug (6.9% vs 13.6%) and a lower rate of dose modification due to toxic effects (16.1% vs 35.2% for CARB and 37.5% for PAX). AEs leading to death occurred in 3.8% of the patients treated with GEF and in 2.7% of the patients treated with PAX + CARB. Interstitial lung disease was significantly more common in patients treated with GEF than in those treated with PAX + CARB or DOC + CIS, including one fatality in each trial. 15,63–65

Table 38 shows the top 10 AEs that occur in the greatest proportion of patients across all arms that use each chemotherapy. The AEs are all grades 3 and 4 (with the exception of one trial58 in Table 38 in which the grades for febrile neutropenia were not specified for either arm); however, reporting of AEs varied (for example grade 3 only, grade 4 only, grade 3 or grade 4 and grade 3 plus grade 4). Certain AEs were grouped together: anaemia haemoglobin was categorised into anaemia; neutrophils to neutropenia; sensory neuropathy, motor neuropathy and neurotoxic effects were all grouped into neuropathy.

Table 38 compares the profile of AEs within each chemotherapy regimen and should not be used to compare toxicities across the different drug regimens. Table 38 shows each drug regimen differs in toxicity profile in terms of percentage of AE.

Table 38 shows that the most common AEs are neutropenia, anaemia and leucopenia. Neutropenia is the top AE for VNB, PAX and DOC and granulocytopenia is the top AE for GEM and PEM. Neutropenia, leucopenia, granulocytopenia all describe a fall in the white blood count and so the common AEs are similar across all the chemotherapy drugs with the exception of GEF, which appears to have a different toxicity profile; the top AE for GEF is aminotransferase elevation. The highest proportion experiencing neutropenia (71%) was among those taking DOC.

Quality of life

Twelve trials15,43–46,48,51,52,55,57,59,62,64 reported QoL outcomes and are listed in Appendix 28. It is surprising, given the importance of QoL, that 1147,49,50,53,54,56,58,60,61,63,65 of the 23 trials do not report QoL data, including three trials60,63,65 that were published in 2010. This could indicate outcome reporting bias, with trial authors failing to present results because they are not statistically significant. QoL was the primary outcome in two trials45,62 and, in the trial by Gridelli et al. ,45 QoL data were assessed according to GEM + VNB compared with PLAT-based chemotherapy (the GEM + VNB combination is not included in this review). Meta-analysis was not performed for QoL data owing to limited data and variability in outcome assessment measures.

A number of instruments/tools that measure QoL were employed in the included trials. The EORTC QLQ-C3029 and the lung cancer-specific module QLQ-LC1330 were used in five trials, the LCSS31 by three trials, and the FACT-L32 questionnaire by three trials. 15,48,57,64

Seven trials48,45,51,52,55,59,62 reported no significant difference in QoL between treatment groups. Four trials15,43,44,46,64 reported some significant differences between treatment groups for QoL; however, in one of these trials,43 results after two cycles of chemotherapy favoured the PAX + CARB arm over the VNB + CIS arm, and results after four cycles favoured the VNB + CIS arm.

In one trial,15,64 significantly more patients in the GEF group than in the PAX + CARB group had a clinically relevant improvement in QoL, as assessed by scores on the FACT-L questionnaire (odds ratio = 1.34; 95% CI 1.06 to 1.69; p = 0.01) and by scores on the Trial Outcome Index (TOI) (which is the sum of the physical well-being, functional well-being and lung cancer subscale scores of FACT-L; odds ratio = 1.78; 95% CI 1.40 to 2.26; p< 0.001).

In another trial46 comparing GEM + CIS with PAX + CIS, no significant difference in global QoL was observed; however, a statistically and clinically significant overall improvement was observed for peripheral neuropathy and alopecia in the GEM + CIS arm compared with the PAX + CIS arm.

Patients treated with DOC + PLAT reported consistently improved global QoL compared with patients treated with VNB + CIS, who generally experienced deterioration in QoL in the trial by Fossella et al. 44

In summary, PAX + PLAT may be associated with worse QoL for alopecia and peripheral neuropathy compared with VNB + PLAT and GEM + PLAT. GEM + PLAT may be associated with better QoL for peripheral neuropathy compared with PAX + PLAT and VNB + PLAT; however, there is a paucity of QoL data available to draw any firm conclusion.

Discussion

Time frame of searching

The electronic searches for the cost-effectiveness review were originally developed for the same time frame as the clinical-effectiveness review (1980–2010); however, it was later decided to include only those trials published after the year 2000 as active chemotherapy treatments for patients with lung cancer have been evolving rapidly since this date. The clinical effectiveness and cost-effectiveness review of lung cancer treatments by Clegg et al. 39 was published in May 2001 and included economic evaluations up to and including 2000. None of the individual studies identified by Clegg et al. 39 are therefore included in this systematic review.

Identification of economic evaluations

A total of 1510 publications were identified as a result of the electronic searches. During stage 1, these studies were screened and duplicated papers were removed. In stage 2, titles and abstracts were screened and 15 papers39,92–105 were selected for potential inclusion in the review. Inclusion and exclusion criteria were applied to these 15 full papers and seven reports39,93–95,97,99,101 were included in the review. The flow diagram in Figure 24 shows the number of reports available at each stage of the inclusion process.

FIGURE 24.

Flow diagram at different stages.

The lung cancer costing model discussed in the two publications by Clegg et al. 39,93 are focused primarily on chemotherapy compared with BSC. However, as they do include two chemotherapy versus chemotherapy comparisons as part of their detailed economic analysis, all data have been extracted and included in this review for information purposes only.

Relevant data were extracted from six evaluations from seven included publications39,93–95,97,99,101 into evidence tables (see Tables 42–45). All data were checked for accuracy by a second reviewer. The eight full-text reports92,96,98,100,102–105 that were excluded during the latter stages of the inclusion process are listed in Table 40 alongside reasons for exclusion. Of these eight trials, five96,102–105 were excluded as they were cost-minimisation analyses (CMAs) only. CMAs were explicitly excluded from the literature review as there are no published results from clinical equivalence trials between chemotherapy regimens for patients with NSCLC in the first-line setting to support such an analysis.

The quality of the reports was assessed using the 35-item list described by Drummond and Jefferson,106 the results of the quality assessment exercise are shown in Table 41. All of the reports are of good/reasonable methodological quality. They typically include the key components of a credible economic evaluation. The key methodological weaknesses include the following: a lack of detail on costs (e.g. no separation of quantity of resources consumed from unit costs); non-explicit statement of length of time horizon or discount rate used; and, in some cases, the authors did not provide disaggregated outcomes or carry out incremental analyses. The main weaknesses of the reports included in the review stem not from their quality but from their limited relevance to UK decision-making. This is a result of the comparisons considered and choice of incremental cost-effectiveness ratio (ICER) rarely being cost per QALY gained.

Study characteristics and model overview

Three97,99,101 of the seven included studies are CEAs. Two papers39,93 are based on the use of three different economic models: a pair-wise comparison between the regimens or BSC (model 1), a CMA (model 2) and a CEA with BSC as the comparator (model 3); only model 1 included a chemotherapy compared with chemotherapy comparison. The study by Klein et al. 95 presents results from both a CMA and a cost–utility analysis (CUA) and the study by Dooms et al. 94 is a CUA.

Klein et al. 95 uses a Markov framework with an initial simple decision tree covering a 6-month period followed by three 6-month cycles. The other studies use simple decision trees with time horizons of ≤ 1 year in four trials,39,93,94,97 3 years101 and 40 months. 99 Most of the economic evaluations have short time frames; this is because they are based on clinical trials with mean OS estimates of approximately 10 months. All of the reports, but one,94 have been conducted using a third-party payer perspective taking account of direct costs only. Dooms et al. 94 adopts a societal perspective using direct costs and costs related to travel expenses. Three39,93,97 are UK based, one is set in Belgium,94 one in Greece,99 one in the Netherlands101 and one in the USA. 95

Four reports39,93,94,99 were funded from public grants or from university funds, two95,97 were funded by a pharmaceutical company and one101 was funded jointly by a pharmaceutical company and several hospitals.

The comparators used in each of the studies and detailed information about design and trial characteristics are presented in Table 42.

Model inputs and data sources

Costs were typically divided into the following categories: costs of drug administration, side effects costs, acquisition costs of drugs, costs of BSC, costs of tests/investigations; and the costs of travel expenses were considered in the only study94 using a societal perspective. The sources included public costs databases, hospital costing data and Medicare reimbursement rates. In general, costs were extracted from publicly available documents, which adds transparency to the costing approaches described in the studies. The economic models, cost item and the sources used are summarised in Tables 43 and 44.

The most commonly used efficacy outcome was survival time with median survival time (MST) used in four39,93,97,101 and OS used in three. 94,97,99 Response rates or ORRs were also used. All efficacy data used in the included studies are shown in Table 45. Sources of efficacy data are varied; a single clinical trial was used in five94,95,97,99,101 of the seven reports and the remaining two39,93 took data from a collection of trials using a mixed-treatment comparison to summarise the data.

Five studies used data from trials39,93–95,99 and used life-year saved (LYS) or LYG as the primary health outcome, whereas two used QALYs. 4,116 Different approaches for calculating QALYs were used in each of the latter of these studies. Klein et al. 95 adopted a CUA approach and used utility values from the published study by Naffes et al. 116 to calculate the QALYs gained in each regimen. Dooms et al. 94 used one item from the LCSS QoL instrument and transformed this into a corresponding utility value; these utilities were then combined with the survival data from a RCT in order to obtain QALYs. Several studies expressed their incremental ratios in terms of cost per progression-free life-year, cost per tumour response or costs to improve mean survival.

Results and sensitivity analysis

Tables 46 and 47 show the cost-effectiveness results, sensitivity analyses and conclusions of the reports. The results of the economic evaluation by Clegg et al. 39,93 reveal that any chemotherapy regimen is cost-effective (vs BSC) at a threshold of £30,000 per LYS except PAX. In the chemotherapy versus chemotherapy comparisons, GEM + CIS dominates GEM and VNB + CIS is cost-effective when compared with VNB. The conclusion of the authors is that depending on the assumptions used, the new drugs range from being cost-effective, as conventionally accepted, to being cost saving. The results of the sensitivity analyses only slightly change the results from the base-case analysis.