First‐line treatment of advanced epidermal growth factor receptor (EGFR) mutation positive non‐squamous non‐small cell lung cancer

Summary of findings for the main comparison. Erlotinib vs control

First‐line treatment of advanced epidermal growth factor receptor (EGFR) mutation positive (M+) non‐squamous non‐small cell lung cancer (NSCLC): erlotinib comparisons
Patient or population: EGFR M+ patients with NSCLC Settings: oncology Intervention: erlotinib Comparison: control (cytotoxic chemotherapy)
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Control	Erlotinib
Overall survival	56 per 100	54 per 100 (46 to 63)	HR 0.95 (0.75, 1.22)	429 (3 studies)	High	All trials were open label but included blinded independent review
Progression‐free survival	73 per 100	33 per 100 (27 to 40)	HR 0.30 (0.24, 0.38)	595 (4 studies)	High	All trials were open label but included blinded independent review
The basis for the assumed risk is calculated as the event rate in the treatment group The corresponding risk is calculated as the assumed risk x the risk ratio (RR) of the intervention where RR = (1 ‐ exp(HR x ln(1 ‐ assumed risk)) )/assumed risk CI:* confidence interval; RR: risk ratio; HR: hazard ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.

Summary of findings 2. Gefitinib vs paclitaxel + carboplatin

First‐line treatment of advanced epidermal growth factor receptor (EGFR) mutation positive (M+) non‐squamous non‐small cell lung cancer (NSCLC): gefitinib comparisons
Patient or population: EGFR M+ patients with NSCLC Settings: oncology Intervention: gefitinib Comparison: paclitaxel + carboplatin
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Paclitaxel + carboplatin	Gefitinib
Overall survival	67 per 100	66 per 100 (58 to 73)	HR 0.95 (0.77 to 1.18)	489 (2 studies)	High	Both trials were open label. IPASS did not report independent blinded review
Progression‐free survival	89 per 100	57 per 100 (50 to 65)	HR 0.39 (0.32 to 0.48)	485 (2 studies)	High	Both trials were open label. IPASS did not report independent blinded review
The basis for the assumed risk is calculated as the event rate in the treatment group The corresponding risk is calculated as the assumed risk x the risk ratio (RR) of the intervention where RR = (1 ‐ exp(HR x ln(1 ‐ assumed risk)) )/assumed risk CI:* confidence interval; RR: risk ratio; HR: hazard ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.

See: Summary of findings for the main comparison Erlotinib vs control; Summary of findings 2 Gefitinib vs paclitaxel + carboplatin; Summary of findings 3 Afatinib vs chemotherapy

Summary of findings 3. Afatinib vs chemotherapy

First‐line treatment of advanced epidermal growth factor receptor (EGFR) mutation positive (M+) non‐squamous non‐small cell lung cancer (NSCLC): afatinib comparisons
Patient or population: EGFR M+ patients with NSCLC Settings: oncology Intervention: afatinib Comparison: cytotoxic chemotherapy
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Cytotoxic chemotherapy	Afatinib
Overall survival	46 per 100	44 per 100 (37 to 52)	HR 0.93 (0.74 to 1.17)	709 (2 studies)	High	Both trials were open label but included blinded independent central review
Progression‐free survival	56 per 100	29 per 100 (24 to 35)	HR 0.42 (0.34 to 0.53)	709 (2 studies)	High	Both trials were open label but included blinded independent central review
The basis for the assumed risk is calculated as the event rate in the treatment group The corresponding risk is calculated as the assumed risk x the risk ratio (RR) of the intervention where RR = (1 ‐ exp(HR x ln(1 ‐ assumed risk)) )/assumed risk CI:* confidence interval; RR: risk ratio; HR: hazard ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.

Background

available in

Description of the condition

Lung cancer is the most common cancer in the world and the second most common cancer diagnosed in the UK (Cancer Research UK). Globally, in 2012, 1.8 million people were diagnosed with lung cancer, representing 12.9% of all cancers (GLOBOCAN 2012). In the UK in 2012, 45,000 new cases of lung cancer were diagnosed, 13% of all new cancers (Cancer Research UK 2012b). In both men and women, smoking is the primary cause of lung cancer (Cancer Research UK 2013). Prognosis is poor, as early‐stage lung cancer is often asymptomatic, and the majority of patients are diagnosed at a late stage. (Cancer Research UK 2012b). In the UK in 2012, 35,000 people died of lung cancer, representing 22% of all deaths from cancer in the UK (Cancer Research UK 2012a).

Non‐small cell lung cancer (NSCLC) accounts for the majority (85% to 90%) of lung cancer cases in the UK and comprises two main histological subgroups: squamous cell carcinoma and non‐squamous cell carcinoma (Cancer Research UK 2012c). Squamous cell carcinoma accounts for 25% to 30% of all NSCLC cases, whilst non‐squamous cell carcinoma (including adenocarcinoma and large cell carcinoma) accounts for 29% of NSCLC cases. Approximately 12% to 13% of patients have NSCLC that is ‘not‐otherwise specified’ with the diagnosis based on cytology alone (NLCA 2015; Schiller 2002). The prognosis for people with NSCLC is poor, with a median survival of the order of six months.

Treatment for people with NSCLC depends not only on the histological subtype and genetic subtype of the tumour, but also on disease stage, comorbidity, and performance status. Chemotherapy, in most cases comprising a cisplatin doublet, for advanced disease can extend overall survival by several months compared to best supportive care and improves quality of life (Brown 2013).

In recent years the biological subtypes of NSCLC have become relevant to the selection of treatment regimens. Attention has been drawn to tumours that harbour the epidermal growth factor receptor mutation (EGFR M+). The EGFR, a protein located on the cell surface, binds to and activates epidermal growth factor. This binding induces receptor dimerisation and tyrosine kinase autophosphorylation, leading through signal transduction to cell proliferation (Han 2012; NCBI). It is estimated that 10% to 15% of people with non‐squamous NSCLC have tumours that are EGFR M+ (Peters 2012; Rosell 2012). An EGFR mutation is more frequently observed in never‐smokers than ever‐smokers (51% versus 10%), in adenocarcinomas compared to cancers of other histologies (40% versus 3%), in people of East Asian ethnicity versus other ethnicities (30% versus 8%), and in females rather than males (42% versus 14%) (Rosell 2009; Scoccianti 2012; Ulivi 2012).

The identification of people with EGFR M+ tumours has led to the development of targeted therapies comprising small molecule tyrosine kinase inhibitors (TKIs) directed at the signal transduction pathway between the cell membrane and the nucleus, while monoclonal antibodies (MABs) bind to and inactivate the receptor on the cell membrane. Since the majority of the phase III trials in this review were started, it has become apparent that activating mutations in exons 19 and 21 are associated with response to the TKIs, while the 1% of tumours with the exon 20 T790M mutation are resistant. The TKIs are orally administered agents, while the MABs are given intravenously. People of interest to this review were chemotherapy‐naive patients with locally advanced or metastatic (stage IIIB or IV) EGFR M+ NSCLC who were not suitable for treatment with curative intent, such as surgery or radical radiotherapy.

Description of the intervention

In Europe, there are three licensed treatments that target EGFR M+ NSCLC: afatinib, erlotinib, and gefitinib. These drugs are TKIs of EGFR and target proteins on the cancer cells related to activation of the signal transduction pathway. These treatments are taken orally (tablets) daily until the disease progresses. Other drugs, for example the TKI dacomitinib and the MAB cetuximab, are currently under clinical investigation and are not yet licensed for the first‐line treatment of people with EGFR M+ NSCLC. We did not assess newer drugs that target the exon 20 T790M mutation in this review.

In the UK, the National Institute for Health and Care Excellence has recommended the use of monotherapy erlotinib, NICE 2012, monotherapy gefitinib, NICE 2010, and more recently, monotherapy afatinib, NICE 2014, for the first‐line treatment of EGFR M+ NSCLC. In Europe, European Society for Medical Oncology guidelines recommend first‐line treatment with monotherapy afatinib, erlotinib, or gefitinib (Reck 2014). In the USA, the Food and Drug Administration has approved the use of monotherapy erlotinib and monotherapy afatinib (FDA 2013; FDA 2014). Globally, there is considerable variation in the use of each of these drugs to treat people with NSCLC and in the availability and quality control of mutation testing, which determines patient selection.

Why it is important to do this review

Treatments for people with NSCLC have been evolving rapidly following the Medical Research Council meta‐analysis that demonstrated improved survival for chemotherapy compared with best supportive care (MRC 1995). Until early 2000, people with NSCLC were offered standard cytotoxic chemotherapy treatments (for example cisplatin, docetaxel, vinorelbine, paclitaxel, and gemcitabine), often given in two‐drug platinum‐based combinations (Brown 2013). However, in recent years patients have been treated with drugs according to their histological subtype (for example pemetrexed plus cisplatin for non‐squamous disease). Even more recently, as understanding of NSCLC has evolved, targeted treatments have been developed to treat specific groups of patients based on molecular criteria, for example TKIs and MABs. It is estimated that around 10% (n = 4000 annually) of all lung cancer patients in the UK have locally advanced or metastatic EGFR M+ NSCLC (NICE 2010), with a higher prevalence in Asian populations. It is therefore important to synthesise evidence for the clinical effectiveness and toxicity of these new treatments to ensure that patients are being treated with the most clinically effective drugs for their specific disease subtype.

Objectives

available in

To assess the clinical effectiveness of single‐agent or combination EGFR therapies used in the first‐line treatment of people with locally advanced or metastatic EGFR M+ NSCLC compared with other cytotoxic chemotherapy agents used alone or in combination, or best supportive care (BSC). The primary outcome was overall survival. Secondary outcomes included progression‐free survival, response rate, toxicity, and quality of life.

Methods

available in

Criteria for considering studies for this review

Types of studies

Parallel randomised controlled trials (RCTs).

Types of participants

Chemotherapy‐naive patients with locally advanced or metastatic (stage IIIB or IV) EGFR M+ NSCLC unsuitable for treatment with curative intent with surgery or radical radiotherapy. We included studies that included or excluded exon 20 T790 in the review.

Types of interventions

EGFR M+ targeted agents, alone or in combination with cytotoxic agents, compared with cytotoxic agents used alone or in combination or BSC.

We excluded trials comparing single‐agent or combinations of cytotoxic chemotherapy without a targeted therapy in either arm and trials with targeted therapy in both arms, and we did not evaluate maintenance or second‐line strategies. We also excluded cross‐over trials.

Types of outcome measures

Primary outcomes

Overall survival

Secondary outcomes

Progression‐free survival
Tumour response
Toxicity and adverse effects of treatment
Quality of life (e.g. Functional Assessment of Cancer Therapy ‐ Lung (FACT‐L) and Trial Outcome Index (TOI))
Symptom palliation

Search methods for identification of studies

Electronic searches

We searched the following electronic databases for relevant published literature up to 1 June 2015. We did not restrict searches by language.

Cochrane Central Register of Controlled Trials (CENTRAL) (2015, Issue 6) Appendix 1
MEDLINE (from 1980) (accessed via PubMed and OvidSP) Appendix 2
EMBASE (from 1946) (OvidSP) Appendix 3
ISI Web of Science (from 1899) Appendix 4

We ran an initial search in October 2012. We ran an updated search (updated by the Cochrane Lung Cancer Group Trials Search Co‐ordinator) in January 2014 and June 2015. As the updated search (Appendix 2) included amendments to the initial search strategy, we conducted a PubMed search from inception to June 2015 to ensure that no relevant articles had been missed. We compared the results of the overall PubMed search with the results of all other searches and examined any non‐duplicate articles for possible inclusion in the review. We identified no relevant publications.

Searching other resources

We searched bibliographies of identified sources and use of Evidence Review Group (ERG) reports to the National Institute for Health and Care Excellence. We searched the proceedings of relevant conferences such as the American Society for Clinical Oncology and the European Society for Medical Oncology up to June 2015. If data were available, we considered including them in the review.

We developed a database of relevant references using EndNote X5 software (Thomson Reuters).

Data collection and analysis

Selection of studies

Two review authors independently took part in all stages of trial selection (FV and VB Search 1; VB and JG Search 2; JAG and YD, JAG and JG Search 3). Review authors first independently scanned the titles and abstracts of references identified by the search strategy. We obtained full details of possibly relevant trials and independently assessed these for inclusion in the review. In case of disagreement, the review authors attempted to reach consensus by discussion, or by involving a third review author (AB or JG). We excluded trials that did not meet all of the inclusion criteria and listed their bibliographic details with reasons for exclusion. We listed ongoing trials that did not report relevant data but met the inclusion criteria for future use. We included trials published in abstract form only if it was clear that the trial was eligible. If it was not clear, we contacted authors for further information and placed the trial in ‘awaiting assessment’ until we received a reply.

Data extraction and management

Two review authors carried out the data extraction (FV and VB Search 1; VB and JG Search 2; JAG and JG Search 3) using pre‐tested data extraction forms, and a third review author (KD) independently checked the extracted data for accuracy. We extracted data relating to the outcome measures as well as information on trial design and participants (for example baseline characteristics). Where data from trials were presented in multiple publications, we extracted and reported these as a single trial with all other relevant publications listed.

Assessment of risk of bias in included studies

We assessed each included trial for risk of bias using criteria outlined in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (see domains listed below) (Higgins 2011). Two review authors (FV and JG Search 1; JG and KD Search 2) independently carried out the assessments. Any disagreements were resolved through discussion.

Random sequence generation (selection bias).
Allocation concealment (selection bias).
Blinding of participants (performance bias).
Blinding of outcome assessment (detection bias).
Incomplete outcome data (attrition bias).
Selective outcome reporting (reporting bias).
Any other identified bias, including inappropriate influence of funders.

We reported bias as either high, low, or unclear (further details of reporting bias are outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011)). We assessed the domains of blinding and incomplete outcome data at the outcome level.

We presented 'Summary of findings' tables with each outcome graded accordingly using the GRADE approach (GRADE Working Group 2004).

Measures of treatment effect

For binary outcomes, where sufficient data were available, we presented relative treatment effects in the form of risk ratio with 95% confidence intervals (CI). For continuous outcomes, we calculated mean differences and 95% CIs provided there was no evidence that the data were subject to skew. If statistical tests used in the original paper were for skewed data, or if median and interquartile ranges were reported, we assumed the data were skewed. We calculated standardised mean differences for quality of life variables where appropriate. For time‐to‐event outcomes, we extracted log hazard ratios (log HR) when available, with 95% CI. If the log HR was not reported, we requested data from authors.

All trials allowed participant cross‐over to another treatment after progression, but no details were provided regarding how this was dealt with in any of the analyses of overall survival (OS).

We considered trials for inclusion in the review that: (1) provided only unplanned, interim findings; and (2) were continuing to recruit participants, but we did not not include these in the meta‐analysis.

Unit of analysis issues

We did not include trials designed as cross‐over trials, as the use of more than one treatment would impact on the assessment of OS (our primary outcome). However, we noted that many of the RCTs included in our review allowed participants from the control arm access to the intervention treatment when their disease progressed; we acknowledge that this limits our assessment of OS.

Dealing with missing data

We contacted authors (and sponsors) of trials for missing data. In cases where authors did not respond, we categorised the studies as 'awaiting classification'.

Assessment of heterogeneity

We assessed statistical heterogeneity between trials visually by inspection of the forest plots and using the Chi² test (P < 0.1 was considered significant due to the low power of the test). We also calculated the I² statistic, which describes the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error (chance). Values of I² range from 0 to 100, with 0 representing no heterogeneity and 100 representing considerable heterogeneity.

For this review:

0% to 29%, heterogeneity might not be important;
30% to 49% may represent moderate heterogeneity;
50% to 74% may represent substantial heterogeneity; and
75% to 100%, considerable heterogeneity.

Assessment of reporting biases

If we identified a sufficient number of trials, we would construct a funnel plot. If asymmetry was present in the funnel plot, we would explore possible causes of bias, such as heterogeneity or outcome reporting bias. As there were not enough trials (at least 10) included in any one meta‐analysis, we did not include funnel plots in this review.

Data synthesis

We have summarised individual trial data in structured tables and as a narrative description. As a major clinical issue is the toxicity of platinum‐based doublet chemotherapy (cytotoxic chemotherapy), we presented subgroups separately with comparators cytotoxic chemotherapy, single‐agent vinorelbine in elderly participants, and placebo. We regarded the combination of an EGFR‐targeted therapy and cytotoxic chemotherapy versus cytotoxic chemotherapy as a separate comparison in view of concerns about interactions between chemotherapy and a tyrosine kinase inhibitor. We combined data for time‐to‐event outcomes using the generic inverse variance method. We used the Mantel‐Haenszel method for dichotomous outcomes. In future versions of this review where data are available, we may combine continuous outcomes using the inverse variance method.

We conducted meta‐analyses using the fixed‐effect model, unless there was substantial heterogeneity (I² > 50%), in which case we used a random‐effects model as a sensitivity analysis. If there was considerable heterogeneity (I² > 75%) we may have combined data, but our conclusions would highlight the amount of heterogeneity present.

Indirect comparisons and network meta‐analysis

We considered that a network meta‐analysis (NMA) was not appropriate because of the different populations across the included trials. We identified other barriers to conducting NMA: some trials reported adjusted analyses, whereas all other trials reported unadjusted analyses and combining these is statistically unsound; participants in all trials were allowed to switch treatment after progression, and we had no information about how this was handled in the analysis for OS. Finally, the Kaplan‐Meier plots shown in the trial reports crossed in four of the trials, indicating that using a Cox proportional hazards model may not be appropriate.

If in future versions of this review we identify trials that compare different interventions that are sufficiently similar in terms of their populations and outcomes, we may make indirect comparisons for competing interventions that have not been compared directly. Multiple‐treatments meta‐analysis (also referred to as network meta‐analysis) may combine direct and indirect comparisons using multivariate meta‐analysis, as this will also take into account any multi‐arm trials. We will use a random‐effects model within STATA to conduct analyses using code from www.mtm.uoi.gr.

We will evaluate transitivity (the trials making different direct comparisons must be sufficiently similar in all respects other than the treatments being compared) clinically. We will compare the distributions of possible effect modifiers (smoking status, age, gender, ethnicity, and performance status) across comparisons using subgroup analysis. As the review only considers first‐line treatment, indications are similar.

We will evaluate consistency using a loop‐specific approach (Salanti 2009), and use a design interaction consistency model (Higgins 2012). If we identify inconsistency, we will not present the network meta‐analysis.

We will assess estimates of treatment effect by pairwise meta‐analysis. We will conduct network meta‐analysis where appropriate.

Prior to analysis we will draw a diagram of the network for all relevant interventions, indicating the number of trials per comparison. We will derive and display ranking probabilities for each treatment using the Surface Under the Cumulative RAnking curve (SUCRA) plot and rankograms (Salanti 2011).

We will discuss the possible effects of risk of bias on the clinical effectiveness data and review findings.

Subgroup analysis and investigation of heterogeneity

In an update of this review when sufficient trials are included and where data are available, we may conduct analyses to investigate any differential effects in terms of:

smoking status
age
sex
ethnicity
performance status
type of mutation (exon 19/exon 21)
type of histology

Sensitivity analysis

In an update of this review when sufficient trials are included, we will conduct sensitivity analyses based on the overall risk of bias of the included trials. We will base overall risk of bias on sequence generation, allocation concealment, and blinding (for the specific outcome), and will base the summary assessment on recommendations in Table 8.7a of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).

Results

Description of studies

Results of the search

The database search strategy yielded 7674 non‐duplicate papers. Of these, we screened 336 full‐text records for inclusion in the review. We identified a further seven records via handsearching of reference lists and found two other records from our search of conference abstracts. We screened all of the potentially relevant references and included 19 eligible RCTs (reported in 56 publications) comparing EGFR‐targeted therapy to chemotherapy as first‐line treatment in NSCLC patients in our review (Figure 1).

Figure 1

Study flow diagram.

We classified three trials as awaiting assessment and have not yet included them in the review (INSPIRE; TALENT; TRIBUTE). We contacted the authors of TALENT and TRIBUTE and asked them to provide data on the EGFR M+ population. We have not received a response. We await the publication of outcomes for the EGFR M+ subgroup from INSPIRE. We found one ongoing trial (ARCHER).

Included studies

See Characteristics of included studies.

The 19 trials that met the inclusion criteria were published or updated between 2003 and 2015 (BMSO99; CHEN; ENSURE; EURTAC; FASTACT 2; First‐SIGNAL; FLEX; GTOWG; INTACT 1; INTACT 2; IPASS; LUX‐Lung 3; LUX‐Lung 6; NEJSG; OPTIMAL; TOPICAL; TORCH; WJTOG3405; Yu 2014). With the exception of GTOWG, all trials were published as peer‐reviewed papers. The overall number of people recruited to the trials ranged between 113, in CHEN, and 1217, in IPASS, with an overall trial population of 9414. The median length of follow‐up (where reported) ranged from 15.9 months, in INTACT 1, to 59 months, in WJTOG3405.

Seven trials included EGFR M+ participants only (ENSURE; EURTAC; LUX‐Lung 3; LUX‐Lung 6; NEJSG; OPTIMAL; WJTOG3405). The number of participants recruited to the EGFR M+ only trials ranged from 165, in OPTIMAL, to 364, in LUX‐Lung 6, with a total population of 1672. The remaining 12 trials recruited a 'mixed' population of participants, that is participants were not selected for inclusion in the trial on the basis of their EGFR mutation status. These latter trials reported results for the subgroup of participants with EGFR M+ mutation status. The numbers of participants reported in these subgroups ranged from 10, in GTOWG, to 261, in IPASS, with a combined total of 645. The combined total of participants with EGFR M+ NSCLC was 2317.

Three trials were conducted exclusively in Europe (EURTAC; GTOWG; TOPICAL); 10 were conducted exclusively in Asia (CHEN; ENSURE; FASTACT 2; First‐SIGNAL; IPASS; LUX‐Lung 6; NEJSG; OPTIMAL; WJTOG3405; Yu 2014); and one was conducted in the USA (BMSO99). The remaining trials were more international, (TORCH), (INTACT 2). LUX‐Lung 3, INTACT 1, and FLEX. The seven trials that recruited exclusively EGFR M+ patients were conducted in Asia, ENSURE, LUX‐Lung 6, NEJSG, OPTIMAL, and WJTOG3405, and Europe, EURTAC, with one international trial (LUX‐Lung 3).

Four of the trials were placebo controlled and double blinded (FASTACT 2; INTACT 1; INTACT 2; TOPICAL); the remainder were specifically reported as being open label or did not report blinding status. In the latter case, we assumed these to be open label due to the nature of the interventions and comparator (that is oral versus intravenous treatments). Three of the 19 included trials were phase II (CHEN; GTOWG; Yu 2014), whilst the others were phase III. Fifteen of the 19 trials were partially or totally funded by a pharmaceutical company (BMSO99; CHEN; ENSURE; EURTAC; FASTACT 2; First‐SIGNAL; FLEX; INTACT 1; INTACT 2; IPASS; LUX‐Lung 3; LUX‐Lung 6; OPTIMAL; TOPICAL; TORCH); the NEJSG and WJTOG3405 trials were funded by scientific groups. The funding source for the GTOWG and Yu 2014 trials was not reported.

Four categories of comparisons for all four agents were described:

targeted agent versus established platinum‐based combinations (e.g. cisplatin or carboplatin and gemcitabine or docetaxel) ‐ the term platinum‐based refers to cisplatin or carboplatin based combinations, both drugs being metabolised to the same active moiety;
targeted agent versus single‐agent chemotherapy drug vinorelbine, for which clinical interest is limited to the elderly population due to its favourable toxicity profile;
cytotoxic chemotherapy with the targeted agent versus chemotherapy alone; and
erlotinib versus placebo.

Population characteristics

All trials provided data for age, sex, performance status, and smoking status except for the INTACT 1, INTACT 2, and GTOWG trials (no details of smoking history). The median age of the overall population of all participants in the included trials ranged from 56 to 77 years; the median age of participants in the EGFR M+ only trials ranged from 56 to 65 years. Two trials only included people aged over 70 years (CHEN; GTOWG), and NEJSG and Yu 2014 only reported mean age. There were more females in nine trials (ENSURE; EURTAC; First‐SIGNAL; IPASS; LUX‐Lung 3; LUX‐Lung 6; NEJSG; OPTIMAL; WJTOG3405), and more males in seven trials (BMSO99; CHEN; FLEX; GTOWG; INTACT 1; INTACT 2; TORCH). The majority of participants were of good performance status (ECOG or WHO 0 or 1). The GTOWG abstract did not report performance status.

It is notable that in all of the trials that recruited EGFR M+ patients only, the proportion of females was greater than males (ENSURE; EURTAC; LUX‐Lung 3; LUX‐Lung 6; NEJSG; OPTIMAL; WJTOG3405).

Interventions

Erlotinib

Eight trials used erlotinib (n = 754 EGFR M+) as the EGFR‐targeted therapy (CHEN; ENSURE; EURTAC; FASTACT 2; GTOWG; OPTIMAL; TOPICAL; TORCH). CHEN and GTOWG used the drug vinorelbine as a single agent or with carboplatin, respectively, in elderly populations. In FASTACT 2, erlotinib was used in combination with a platinum doublet containing gemcitabine. We classified trials using erlotinib into the following comparison groups.

Erlotinib versus platinum‐based chemotherapy: One trial compared erlotinib versus gemcitabine plus carboplatin (OPTIMAL), two trials compared erlotinib versus gemcitabine plus cisplatin (ENSURE; TORCH), and one trial compared erlotinib versus docetaxel plus cisplatin or gemcitabine plus cisplatin (EURTAC).
Erlotinib versus vinorelbine +/‐ chemotherapy: One trial compared erlotinib versus vinorelbine (CHEN), one trial compared erlotinib versus carboplatin plus vinorelbine (GTOWG).
Erlotinib plus chemotherapy versus chemotherapy plus placebo: One trial compared erlotinib plus gemcitabine plus carboplatin or cisplatin versus gemcitabine plus carboplatin or cisplatin plus placebo (FASTACT 2).
Erlotinib versus placebo: One trial considered this comparison (TOPICAL).

Gefitinib

Seven trials used gefitinib (n = 773 EGFR M+) as the EGFR‐targeted therapy (First‐SIGNAL; INTACT 1; INTACT 2; IPASS; NEJSG; WJTOG3405; Yu 2014). Three trials used gefitinib in combination with chemotherapy (INTACT 1; INTACT 2; Yu 2014). We classified trials using gefitinib into the following comparison groups.

Gefitinib versus gemcitabine plus cisplatin: One trial considered this comparison (First‐SIGNAL).
Gefitinib versus paclitaxel plus carboplatin: Two trials considered this comparison (IPASS; NEJSG).
Gefitinib versus docetaxel plus cisplatin: One trial considered this comparison (WJTOG3405).
Gefitinib and carboplatin plus paclitaxel or cisplatin plus gemcitabine versus cytotoxic chemotherapy alone: Two trials considered this comparison (INTACT 1; INTACT 2). However, as EGFR M+ specific data from both trials was analysed as though from one trial, and data were only presented narratively.
Gefitinib plus pemetrexed and cisplatin versus pemetrexed plus cisplatin: One trial considered this comparison (Yu 2014).

Afatinib

Two trials compared afatinib (n = 709) with cytotoxic chemotherapy (LUX‐Lung 3; LUX‐Lung 6). These trials differed principally in the selection of the cytotoxic chemotherapy comparator, LUX‐Lung 3 comparing afatinib with cisplatin and pemetrexed in an ethnically diverse population, and LUX‐Lung 6 comparing afatinib with cisplatin and gemcitabine in an Asian population. We combined these trials in a meta‐analysis for progression‐free survival, overall survival, and response.

Cetuximab

Two trials (n = 81) compared cetuximab plus chemotherapy with combination chemotherapy (BMSO99; FLEX).

Of the seven trials that recruited only people with EGFR M+ NSCLC, two trials used afatinib (LUX‐Lung 3; LUX‐Lung 6), three used erlotinib (ENSURE: EURTAC; OPTIMAL), and two used gefitinib (NEJSG; WJTOG3405). All seven EGFR M+ only trials compared targeted treatment with cytotoxic chemotherapy (ENSURE; EURTAC; LUX‐Lung 3; LUX‐Lung 6; NEJSG; OPTIMAL; WJTOG3405).

Outcomes

The primary outcome for the majority of trials was progression‐free survival with secondary outcomes of overall survival, tumour response rate, symptom palliation, quality of life, and safety. Overall survival was the primary outcome in six trials (First‐SIGNAL; FLEX; INTACT 1; INTACT 2; TOPICAL; TORCH).

Excluded studies

See Characteristics of excluded studies.

We excluded 280 records after the selection procedure (Figure 1). The main reasons for exclusion were the use of non‐randomised designs (including systematic reviews and reports from conferences), non‐assessment of participants' EGFR mutation status, and non‐administration of treatments as first‐line therapy. We excluded other trials if they were designed to assess maintenance treatment, or if an EGFR‐targeted therapy was used in both trial arms. We were unable to easily exclude articles at the screening stage, as we could not be certain from the abstract whether subgroup analyses of outcomes of participants with EGFR M+ tumours were reported. In the Characteristics of excluded studies table we have listed the 20 trials that appeared to meet the inclusion criteria, but on closer examination were not a complete match. Participants in five trials were not tested for EGFR mutations (Crino 2008; Gatzemeier 2003; Goss 2009; Lilenbaum 2008; Rosell 2004). Two trials tested for EGFR expression only (Rosell 2008; Thatcher 2014). Three trials included too few participants with EGFR M+ tumours to warrant analysis (FASTACT; Heigener 2014; White), and in eight trials tyrosine kinase inhibitors treatment was included in both trial arms (Hirsh 2011; Janne 2012; JO25567; Massuti 2014; NEJ005 2014; NEJ009; Xie 2015; Yang 2015). One trial only assessed outcomes of patients who had survived at one year (Boutsikou 2013), and in another trial there were insufficient samples available for testing (ECOG 4508).

Risk of bias in included studies

Allocation

Of the 19 included trials, 11 reported adequate information about the methods used to generate the randomisation sequence and the allocation concealment procedure; we considered these trials to be at low risk of bias (EURTAC; FASTACT 2; FLEX; IPASS; LUX‐Lung 3; LUX‐Lung 6; NEJSG; OPTIMAL; TOPICAL; TORCH; WJTOG3405). We considered the risk of bias for the remaining eight trials to be unclear due to lack of reported information (BMSO99; CHEN; ENSURE; First‐SIGNAL; GTOWG; INTACT 1; INTACT 2; Yu 2014).

Blinding

Performance bias

Only 4 of the 19 included trials reported employing blinding procedures (INTACT 1; INTACT 2; NEJSG; TOPICAL). The remaining trials explicitly stated they were open label or did not report blinding status. In the latter case, we assumed these trials were open label due to the differences between interventions and comparator (that is oral versus intravenous).

Detection bias

We considered 11 of the trials to be at low risk of detection bias for the outcome of progression‐free survival, as they incorporated independent verification procedures, in BMSO99, ENSURE, EURTAC, FASTACT 2, First‐SIGNAL, LUX‐Lung 3, LUX‐Lung 6, and NEJSG, or blinded outcome assessment, in INTACT 1, INTACT 2, and TOPICAL. None of the remaining trials reported any independent assessment procedures and were considered to be at high risk of bias for the outcome of progression‐free survival.

Incomplete outcome data

In all trials, all participants were accounted for in the analyses. There did not appear to be any major imbalances in drop‐out rates between trial arms in any of the trials, therefore we considered all trials to be at low risk of bias.

Selective reporting

We considered only one trial to be at high risk of reporting bias (CHEN). The trial protocol stated time to progression as a secondary outcome of the trial, however the published paper did not report this outcome. We considered two trials to be at unclear risk of bias as the available information was insufficient to judge selective reporting (FLEX; GTOWG). We considered all other trials to be at a low risk of bias, as either trial protocols were available, or all outcomes stated in the methods section of the papers were reported.

Other potential sources of bias

Fifteen trials were sponsored fully or in part by pharmaceutical companies. One trial was terminated early as the non‐inferiority of the intervention arm was demonstrated by the first planned interim analysis (TORCH). Two trials were terminated early for benefit (ENSURE; EURTAC).

Effects of interventions

Pairwise meta‐analysis

Erlotinib versus placebo, platinum‐based chemotherapy, or other cytotoxic agents

Primary outcome: Overall survival

Data from five trials were available for overall survival (OS) (CHEN; ENSURE; EURTAC; FASTACT 2; TORCH). Three trials presented limited data (OPTIMAL; TOPICAL), and one trial presented no data (GTOWG).

Erlotinib versus platinum‐based chemotherapy: The pooled treatment effect estimate for three trials, hazard ratio (HR) of 0.95 (95% confidence interval (CI) 0.75 to 1.22; I² = 0; 71%) indicated no significant difference in OS between the groups (ENSURE; EURTAC; TORCH). OPTIMAL reported that OS did not differ significantly between the two treatment arms (HR = 1.065, P = 0.6849). No standard error was reported, so the results could not be entered into a meta‐analysis.

Erlotinib versus vinorelbine: CHEN reported a HR of 2.16 (95% CI 0.58 to 8.10) for OS comparing erlotinib versus vinorelbine in elderly patients, indicating no significant difference in OS between the groups.

Erlotinib plus cytotoxic chemotherapy versus cytotoxic chemotherapy plus placebo: FASTACT 2 reported a HR of 0.48 (95% CI 0.27 to 0.85) for OS indicating a significant difference in OS favouring erlotinib plus cytotoxic chemotherapy in a trial of 91 participants (Analysis 1.1).

Erlotinib versus placebo: TOPICAL reported the median overall survival, which was 10.4 months (95% CI 5.5 to 15.1) for erlotinib (n = 17) versus 3.7 months (95% CI 0.3 to 49.3) for placebo (n = 11).

Secondary outcomes

1. Progression‐free survival

Six trials reported progression‐free survival (PFS) (CHEN; ENSURE; EURTAC; FASTACT 2; OPTIMAL; TORCH). One trial did not report hazard ratios and only presented limited data (TOPICAL), and one trial reported no data (GTOWG).

Erlotinib versus chemotherapy: The pooled treatment effect estimate for four trials (HR 0.30, 95% CI 0.24 to 0.38; fixed‐effect; I² = 74%) favoured erlotinib (ENSURE; EURTAC; OPTIMAL; TORCH). As there was a substantial amount of heterogeneity, we performed a sensitivity analysis using the random‐effects model, and results were similar to the main analysis (HR 0.31, 95% CI 0.20 to 0.50).

Erlotinib versus vinorelbine: CHEN reported a HR of 0.55 (95% CI 0.21 to 1.46) for PFS indicating no significant difference between the groups.

Erlotinib plus cytotoxic chemotherapy versus cytotoxic chemotherapy plus placebo: FASTACT 2 reported a significant difference in PFS favouring erlotinib plus cytotoxic chemotherapy (HR 0.25, 95% CI 0.16 to 0.39) (Analysis 1.2).

Erlotinib versus placebo: TOPICAL reported the median PFS, which was 4.8 months (95% CI 1.6 to 8.8) for erlotinib (n = 17) and 2.9 months (95% CI 0.3 to 10.1) for placebo (n = 11).

ENSURE, EURTAC, and OPTIMAL showed an improvement in PFS for the exon 19 deletion in favour of erlotinib. We did not perform meta‐analysis of this preliminary data.

2. Tumour response

Erlotinib versus platinum‐based chemotherapy: The pooled treatment effect estimate for five trials favoured erlotinib (risk ratio (RR) 2.26, 95% CI 1.85 to 2.76; I² = 57%) (ENSURE; EURTAC; GTOWG; OPTIMAL; TORCH). As there was a substantial amount of heterogeneity, we performed a sensitivity analysis using a random‐effects model, and results were similar (RR 2.20, 95% CI 1.53 to 3.17) (Analysis 1.3).

Erlotinib versus vinorelbine: CHEN reported a RR of 0.83 (95% CI 0.19 to 3.67; 24 participants) for tumour response, indicating no significant differences in tumour response between the groups.

FASTACT 2 observed an objective response in 41 (84%) of 49 participants with EGFR‐activating mutations in the erlotinib plus cytotoxic chemotherapy group, and 7 (15%) of 48 participants in the chemotherapy plus placebo group (RR 5.74, 95% CI 2.86 to 11.50).

TOPICAL did not report tumour response for EGFR M+ participants.

3. Toxicity and adverse effects of treatment.

The most commonly reported adverse effects of treatment (AEs) in participants treated with erlotinib as a monotherapy were rash, diarrhoea, and fatigue (CHEN; ENSURE; EURTAC; GTOWG; OPTIMAL; TOPICAL; TORCH) (Table 1). Other AEs included mouth ulcers, constitutional symptoms, nausea, increased alanine aminotransferase , dyspnoea, and pulmonary toxicities. Cytotoxic chemotherapy was associated with greater grade 3/4 myelosuppression, fatigue (two trials) and anorexia (one trial). Commonly reported AEs in the trial that administered erlotinib in combination with cytotoxic chemotherapy were neutropenia, thrombocytopenia, and anorexia (FASTACT 2).

Table 1. Adverse events ‐ most commonly occurring grade 3 & 4

Study

Definition of AE

Population

Top AE (listed according to intervention)

Second top AE (listed according to intervention)

Third top AE (listed according to intervention)

Top 3 AEs (listed according to comparator)

Afatinib trials

LUX‐Lung 3

Grade >= 3 CTC (V3)

AEs that were reported in > 10% of participants in either group and if there was a >= 10% difference between the groups

EGFR M+ only

Rash/acne:

16.2% (AFA) vs 0% (cytotoxic chemotherapy)

Diarrhoea:

14.4% (AFA) vs 0% (cytotoxic chemotherapy)

Paronychia:

11.4% (AFA) vs 0% (cytotoxic chemotherapy)

Neutropenia: 18% vs 0.4%

Fatigue: 12.6% vs 1.3%

Leukopenia: 8.1% vs 0.4%

LUX‐Lung 6

CTC (V3)

Events are included if reported for >= 1% of participants in any treatment group

EGFR M+ only

Rash/acne:

14.6% (AFA) vs 0% (cytotoxic chemotherapy)

Diarrhoea:

5.4% (AFA) vs 0% (cytotoxic chemotherapy)

Stomatitis/mucositis:

5.4% (AFA) vs 0% (cytotoxic chemotherapy)

Neutropenia: 26.5% vs 0.4%

Vomiting: 19.4% vs 0.8%

Leukopenia: 15.1% vs 0.4%

Erlotinib trials

CHEN

Incidence rate >= 10%

Unselected population

Rash:

64.9% (ERL) vs NR (cytotoxic chemotherapy)

Diarrhoea:

29.8% (ERL) vs NR (cytotoxic chemotherapy)

Mouth ulceration:

14% (ERL) vs NR (cytotoxic chemotherapy)

Anorexia: 26.3% vs NR

Diarrhoea: 12.3% vs NR

Vomiting: 10.5% vs NR

ENSURE

Grade ≥ 3

≥ 5% in either arm

EGFR M+ only

Rash:

6.4% (ERL) vs 1% (cytotoxic chemotherapy)

Neutropenia, leukopenia,

anaemia:

All 0.9% (ERL) vs 25%, 14.4%, 12.5% respectively (cytotoxic chemotherapy)

‐

Neutropenia: 25% vs 0.9%

Leukopenia: 14.4% vs 0.9%

Anaemia: 12.5% vs 0.9%

EURTAC

Grade 3/4 CTC (V3)

Common AEs

EGFR M+ only

Rash:

13% (ERL) vs 0% (cytotoxic chemotherapy)

Fatigue:

6% (ERL) vs 20% (cytotoxic chemotherapy)

Diarrhoea:

5% (ERL) vs 0% (cytotoxic chemotherapy)

Neutropenia: 22% vs 0%

Fatigue: 20% vs 6%

Thrombocytopenia: 14% vs 0%

FASTACT 2

Grade 3/4 CTC (V3)

Most commonly reported

Unselected population

Neutropenia:

29% (ERL) vs 25% (cytotoxic chemotherapy)

Thrombocytopenia

14% (ERL) vs 14% (cytotoxic chemotherapy)

Anaemia:

11% (ERL) vs 9% (cytotoxic chemotherapy)

Neutropenia: 25% vs 29%

Thrombocytopenia: 14% vs 14%

Anaemia: 9% vs 11%

GTOWG

Grade 3/4

Unselected population

Rash:

12% (ERL) vs 0% (cytotoxic chemotherapy)

Diarrhoea:

6% (ERL) vs 2% (cytotoxic chemotherapy)

Constitutional symptoms:

3% (ERL) vs 5% (cytotoxic chemotherapy)

Neutropenia: 36% vs 0%

Leukocytes: 33% vs 0%

Haemoglobin: 11% vs 0.7%

OPTIMAL

Grade 3/4 CTC (V3)

AEs occurred in 3% or more in either treatment group

EGFR M+ only

Increased ALT:

4% (ERL) vs 1% (cytotoxic chemotherapy)

Skin rash:

2% (ERL) vs 0% (cytotoxic chemotherapy)

Diarrhoea:

1% (ERL) vs 0% (cytotoxic chemotherapy)

Neutropenia: 42% vs 0%

Thrombocytopenia: 40% vs 0%

Anaemia: 13% vs 0%

TOPICAL

CTC (V3)

Specific AEs grade 3 or 4

Unselected population

Dyspnoea:

59% (ERL) vs 64% (PLA)

Fatigue:

23% (ERL) vs 23% (PLA)

Diarrhoea:

8% (ERL) vs 1% (cytotoxic chemotherapy)

Dyspnoea:

64% vs 59%

Fatigue:

23% vs 23%

Anorexia: 5% vs 5%

TORCH

Worst toxicity experienced with first‐line treatment alone

Unselected population

Skin rash:

11% (ERL) vs 0% (cytotoxic chemotherapy)

Pulmonary toxicity:

9% (ERL) vs 6% (cytotoxic chemotherapy)

Fatigue:

8% (ERL) vs 12% (cytotoxic chemotherapy)

Neutropenia: 21% vs 0%

Thrombocytopenia: 12% vs 0%

Fatigue: 12% vs 8%

Gefitinib trials

First‐SIGNAL

Grade 3 or 4 CTC (V3)

Unselected

population

Rash:

29.3% (GEF) vs 2% (cytotoxic chemotherapy)

Anorexia:

13.8% (GEF) vs 57.3% (cytotoxic chemotherapy)

AST:

11.3% (GEF) vs 2% (cytotoxic chemotherapy)

Anorexia: 57.3% vs 13.9%

Neutropenia: 54% vs 1.9%

Fatigue: 45.3% vs 10.1%

INTACT 1

Grade 3/4 CTC

Commonly occurring AEs

Unselected

population

Thrombocytopenia*:

5.8% (GEF + cytotoxic chemotherapy) vs 5.6% (cytotoxic chemotherapy)

Rash:

3.6% (GEF + cytotoxic chemotherapy) vs 1.1% (cytotoxic chemotherapy)

Diarrhoea:

3.6% (GEF + cytotoxic chemotherapy) vs 2.3% (cytotoxic chemotherapy)

Thrombocytopenia*: 5.6% vs 5.8%

Leukopenia: 2.5% vs 3.3%

Diarrhoea: 2.3% vs 3.6%

INTACT 2

Grade 3/4 CTC (V2)

Common drug‐related AEs

Unselected

population

Diarrhoea:

9.9% (GEF + cytotoxic chemotherapy) vs 2.9% (cytotoxic chemotherapy)

Neutropenia:

6.7% (GEF + cytotoxic chemotherapy) vs 5.9% (cytotoxic chemotherapy)

Rash:

3.2% (GEF + cytotoxic chemotherapy) vs 1.5% (cytotoxic chemotherapy)

Neutropenia: 5.9% vs 6.7%

Diarrhoea: 2.9% vs 9.9%

Vomiting: 2.3% vs 2%

IPASS

Grade 3, 4, or 5 CTC (V3)

At least 10% of participants in either treatment group and at least a 5% difference between arms

Unselected

population

Diarrhoea:

3.8% (GEF) vs 1.4% (cytotoxic chemotherapy)

Any neutropenia:

3.7% (GEF) vs 67.1% (cytotoxic chemotherapy)

Rash:

3.1% (GEF) vs 0.8% (cytotoxic chemotherapy)

Any neutropenia: 67.1% vs 3.7%

Leukopenia: 35% vs 1.5%

Anaemia: 10.6% vs 2.2%

NEJSG

Grade >= 3 CTC (V3)

At least 10% of participants in either treatment group and at least a 5% difference between arms

EGFR M+ only

ATE:

26.3% (GEF) vs 0.9% (cytotoxic chemotherapy)

Rash:

5.3% (GEF) vs 2.7% (cytotoxic chemotherapy)

Appetite loss:

5.3% (GEF) vs 6.2% (cytotoxic chemotherapy)

Neutropenia: 65.5% vs 0.9%

Arthralgia: 7.1% vs 0.9%

Neuropathy: 6.2% vs 0%

Appetite loss: 6.2% vs 5.3%

WJTOG3405

Grade >= 3 CTC (V3)

AEs occurred in 10% of either of the treatment groups

EGFR M+ only

ALT/AST:

27.5% (GEF) vs 2.3% (cytotoxic chemotherapy)

Rash:

2.3% (GEF) vs 0% (cytotoxic chemotherapy)

Fatigue:

2.3% (GEF) vs 2.3% (cytotoxic chemotherapy)

Neutropenia: 84% vs 0%

Leucocytopenia: 50% vs 0%

Anaemia: 17% vs 0%

Yu 2014

Grade 3+

Participants with at least 1 AE

Unselected

population

Rash:

16% (GEF + cytotoxic chemotherapy) vs 0% (cytotoxic chemotherapy)

Vomiting:

10% (GEF) vs 8% (cytotoxic chemotherapy)

Neutropenia:

10% (GEF) vs 12% (cytotoxic chemotherapy)

Neutropenia: 12% vs 10%

Nausea: 8% vs 5%

Vomiting: 8% vs 10%

Cetuximab trials

BMSO99

Grade 3/4 CTC (V3)

Most frequent and relevant grade 3/4 AEs

Unselected population

Neutropenia:

62.5% (CET + cytotoxic chemotherapy) vs 56% (cytotoxic chemotherapy)

Leukopenia:

43.8% (CET + cytotoxic chemotherapy) vs 30.7% (cytotoxic chemotherapy)

Fatigue:

15.1% (CET + cytotoxic chemotherapy) vs 12.2% (cytotoxic chemotherapy)

Same AEs as intervention

FLEX

Grade 3/4 CTC (V2)

AEs that were reported in > 5% of participants (G3/G4) or > 1% (G4) or AEs of special interest in either group

EGFR M+ expressing

Neutropenia:

53% (CET + cytotoxic chemotherapy) vs 51% (cytotoxic chemotherapy)

Leukopenia:

25% (CET + cytotoxic chemotherapy) vs 19% (cytotoxic chemotherapy)

Febrile neutropenia:

22% (CET + cytotoxic chemotherapy) vs 15% (cytotoxic chemotherapy)

Neutropenia: 52% (cytotoxic chemotherapy) vs 52% CET + cytotoxic chemotherapy

Leukopenia: 19% (cytotoxic chemotherapy) vs 25% (CET vs cytotoxic chemotherapy)

Anaemia: 16% (cytotoxic chemotherapy) vs 1% (CET + cytotoxic chemotherapy)

AE: adverse event
AFA: afatinib
ATE: aminotransferase elevation
ALT: alanine aminotransferase
AST: aspartate aminotransferase
CET: cetuximab
CTC: common toxicity criteria
ERL: erlotinib
EGFR M+: epidermal growth factor receptor mutation positive
GEF: gefitinib
NR: not reported

PLA: placebo

*Neutropenia was also reported as 5.8% for G3/4; as this rate was higher than the rate for all participants (5%) it was not included in the table.

4. Quality of life

Two trials reported on the quality of life (QoL) of EGFR M+ participants (OPTIMAL; TORCH). One trial used the Lung Cancer Symptom Scale (LCSS) to measure QoL, but compliance was so poor that the authors regarded the analysis as inconclusive (EURTAC).

QoL was measured but not reported in the trial reports in GTOWG, and was not available for the EGFR M+ subgroup in three trials (CHEN; FASTACT 2; TOPICAL).

TORCH used the the European Organisation for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire ‐ Core 30 (QLQ‐C30) and the lung cancer‐specific module (EORTC QLQ‐LC13) to evaluate QoL. The number of participants improved/stable/worse was reported for selected and unselected participants receiving erlotinib and chemotherapy. Improvement in terms of global QoL and physical functioning was particularly evident in the small numbers of EGFR M+ participants (n = 36/39 available for analysis) for erlotinib compared to cytotoxic chemotherapy.

OPTIMAL used the Functional Assessment of Cancer Therapy‐Lung (FACT‐L), LCSS, and Trial Outcome Index (TOI) to assess QoL. The odds ratios (ORs) (with covariates EGFR mutation type, smoking history, and histological type) were in favour of erlotinib and were 6.69 (95% CI 3.01 to 14.85; P = 0.0001), 7.54 (95% CI 3.38 to 16.85; P = 0.0001), and 8.07 (95% CI 3.57 to 18.26; P = 0.0001), respectively.

In the ENSURE trial, deterioration in TOI was 11.4 months for erlotinib compared to 4.2 months for chemotherapy (HR 0.51, 95% CI 0.34 to 0.76; P = 0.0006), and time to deterioration in QoL was 8.2 months for erlotinib compared to 2.8 months for chemotherapy (HR 0.64, 95% CI 0.44 to 0.93; P = 0.0168).

5. Symptom palliation

In the TORCH trial, the time to deterioration curves for cough, dyspnoea, and pain in the first 20 weeks were visually assessed for erlotinib versus chemotherapy, and no major differences were observed. No statistical analyses were provided by the authors.

The OPTIMAL trial reported that the time to improvement of symptoms on the FACT‐L, TOI, and LCSS (sometimes abbreviated to Lung Cancer Subscale (LCSS)) was significantly shorter for erlotinib compared to chemotherapy: FACT‐L 1.51 versus 3.19 months (P = 0.0067); TOI 2.79 versus 3.48 months (P = 0.003); LCSS 1.48 versus 3.15 months (P = 0.0010). There was also significant correlation between overall response and improvement in symptom scores (P = 0.0006, 0.0002, and 0.0213 for FACT‐L, TOI, and LCSS, respectively).

In the ENSURE trial, preliminary data using the FACT‐L showed that time to symptomatic progression was 13.8 months for erlotinib compared to 5.5 months for chemotherapy (HR 0.56, 95% CI 0.36 to 0.87; P = 0.0076).

Gefitinib versus cytotoxic chemotherapy

Primary outcome: Overall survival

We could not combine data for all four trials comparing gefitinib to platinum‐based chemotherapy (First‐SIGNAL; IPASS; NEJSG; WJTOG3405), as two trials reported only adjusted analyses (IPASS; NEJSG). It is not advisable to combine adjusted and unadjusted estimates.

Gefitinib versus gemcitabine plus cisplatin: One trial, First‐SIGNAL, reported a HR of 1.04 (95% CI 0.50 to 2.20)

Gefitinib versus carboplatin and paclitaxel: Pooled analysis of the two trials indicated no significant difference in OS between the groups (HR 0.95, 95% CI 0.77 to 1.18; I² = 0) (IPASS; NEJSG).

Gefitinib versus docetaxel plus cisplatin: WJTOG3405 reported a HR of 1.25 (95% CI 0.88 to 1.78), indicating no significant difference in OS between the groups (Analysis 2.1).

Gefitinib and platinum‐based chemotherapy versus platinum‐based chemotherapy. INTACT 1 and INTACT 2 reported a combined HR of 1.77 (95% CI 0.50 to 6.23), indicating no significant difference in OS between the groups. Yu 2014 did not report on OS.

Secondary outcomes

1. Progression‐free survival

Gefitinib versus gemcitabine plus cisplatin: First‐SIGNAL reported a HR of 0.54 (95% CI 0.27 to 1.10), indicating no significant difference in PFS between the groups.

Gefitinib versus paclitaxel plus carboplatin: The pooled treatment effect estimate for two trials showed a significant difference in PFS between the groups, favouring gefitinib (HR 0.39, 95% CI 0.32 to 0.48; I² = 73%) (IPASS; NEJSG). As there was a substantial amount of heterogeneity, we performed a sensitivity analysis using a random‐effects model, and results were similar (HR 0.39, 95% CI 0.26 to 0.59).

Gefitinib versus docetaxel plus cisplatin: WJTOG3405 reported a significant difference in PFS favouring gefitinib (HR 0.49, 95% CI 0.34 to 0.71) (Analysis 2.2).

Gefitinib and cytotoxic chemotherapy versus cytotoxic chemotherapy: INTACT 1 and INTACT 2 reported a HR of 0.55 (95% CI 0.19 to 1.60), indicating no significant difference in PFS between the groups in a combined total of 32 participants.

Yu 2014 reported a HR of 0.20 (95% CI 0.05 to 0.75) for PFS for comparison of gefitinib plus pemetrexed and cisplatin vs pemetrexed plus cisplatin.

IPASS and NEJSG both showed an improvement in PFS for the exon 19 deletion in the gefitinib population.

2. Tumour response

The pooled treatment effect estimate for four trials, First‐SIGNAL, IPASS, NEJSG, and WJTOG3405, favoured gefitinib (RR 1.87, 95% CI 1.60 to 2.19; I² = 58%) (Analysis 2.3). As there was a substantial amount of heterogeneity, we performed a sensitivity analysis using a random‐effects model, and results were similar (RR 1.92, 95% CI 1.46 to 2.52).

INTACT 1 and INTACT 2 showed the response rates for gefitinib plus cytotoxic chemotherapy were the same as for cytotoxic chemotherapy alone (30.4% versus 28.7%). Yu 2014 reported a response rate of 77% for cytotoxic chemotherapy plus gefitinib compared to cytotoxic chemotherapy alone (P = 0.13).

Response at cross‐over after progression on first‐line treatment

NEJSG reported that 28.2% of 52 participants responded to carboplatin and paclitaxel after progressing on gefitinib, and 58.5% of 106 participants responded to gefitinib after progressing on carboplatin and paclitaxel.

INTACT 1 and INTACT 2 reported that 13 out of 18 (72%) of EGFR M+ participants responded to gefitinib plus cytotoxic chemotherapy, while 2 out of 5 (40%) of EGFR M+ participants responded to cytotoxic chemotherapy alone.

3. Toxicity and adverse effects of treatment

The most commonly reported AE for gefitinib monotherapy was rash, followed by liver toxicity, anorexia, and diarrhoea (First‐SIGNAL; IPASS; NEJSG; WJTOG3405) (Table 1). Cytoxic chemotherapy was associated with greater grade 3/4 myelosuppression in all comparisons and greater anorexia in one trial. The commonly reported AEs for gefitinib plus cytotoxic chemotherapy were thrombocytopenia, rash, diarrhoea and neutropenia (INTACT 1; INTACT 2).

4. Quality of life

Two trials reported on QoL (IPASS; NEJSG). QoL was measured but not reported in the trial reports in one trial (INTACT 2), not measured in two trials (INTACT 1; WJTOG3405), and not available for the EGFR M+ subgroup in one trial (First‐SIGNAL).

IPASS used the FACT‐L and TOI symptom improvement by the LCSS and achieved 89.5% compliance for the cytotoxic chemotherapy group and 94.8% for the gefitinib group. Gefitinib was significantly favoured over carboplatin plus paclitaxel in the proportion of participants showing improvement in FACT‐L total score, TOI, and LCSS (FACT‐L total score 70.2% versus 44.5% (OR 3.01, 95% CI 1.79 to 5.07), TOI 70.2% versus 38.3% (OR 3.96, 95% CI 2.33 to 6.71), LCSS 75.6% versus 53.9% (OR 2.70, 95% CI 1.58 to 4.62)). The time‐to‐deterioration data showed a median of 15.6 months for gefitinib compared to 3.0 months for cytotoxic chemotherapy for FACT‐L; 16.6 months for gefitinib compared to 2.9 months for cytotoxic chemotherapy for TOI; and 11.3 months for gefitinib compared to 2.9 months for cytotoxic chemotherapy for LCSS. In the 131 participants in the gefitinib group who improved, the median time to improvement in all three scores was either 8 or 11 days.

NEJSG assessed QoL weekly using the Care Notebook and achieved compliance in 72 participants (63%) on chemotherapy and 76 participants (69%) on gefitinib. They used three categories of physical, mental, and "life" well‐being, each of which had three subcategories. The number of participants improved/stable/worse was also reported, and there was no difference between the treatment arms in mental well‐being. However, the physical and life scales were all better for gefitinib than for cytotoxic chemotherapy. The data for daily functioning was quoted as HR 0.32 (95% CI 0.17 to 0.59; P<0001).

5. Symptom palliation

In the NEJSG trial, participants who received gefitinib had a significantly longer time to deterioration up to 20 weeks than participants who received paclitaxel plus carboplatin using both 9.1% and 27.3% levels of deterioration. The data for 27.3% deterioration for pain and shortness of breath showed HR 0.28 (95% CI 0.17 to 0.46; P = 0.0001) in favour of gefitinib.

Afatinib versus cisplatin‐based chemotherapy

Afatinib versus pemetrexed plus cisplatin: One trial considered this comparison (LUX‐Lung 3).

Afatinib versus gemcitabine plus cisplatin: One trial considered this comparison (LUX‐Lung 6).

Primary outcome: Overall survival

The pooled treatment effect estimate indicated no significant difference in OS between the groups (HR 0.93, 95% CI 0.74 to 1.17; I² = 0; 2 trials) (Analysis 3.1), although data for LUX‐Lung 6 were immature. A preliminary report of a pooled analysis of participants with an exon 19 deletion or L858R mutation showed improved survival for afatinib compared to cytotoxic chemotherapy in participants with an exon 19 deletion (HR 0.81, 95% CI 0.66 to 0.99; P= 0.037) (Yang 2014). We did not formally assess analysis of mutation site in this review.

Secondary outcomes

1. Progression‐free survival

The pooled treatment effect estimate showed a significant difference in PFS between the groups favouring afatinib (HR 0.42, 95% CI 0.34 to 0.53; I² = 90%; 2 trials) (Analysis 3.2). As there was a substantial amount of heterogeneity, we performed a sensitivity analysis using a random‐effects model, and results were similar (HR 0.41, 95% CI 0.20 to 0.83).

2. Tumour response

The pooled treatment effect estimate favoured afatinib (RR 2.71, 95% CI 2.12 to 3.46; I² = 0%; 2 trials) (Analysis 3.3).

3. Toxicity and adverse effects of treatment

The most commonly reported grade 3/4 AEs in the afatinib‐treated participants were rash and diarrhoea, paronychia, and stomatitis/mucositis (LUX‐Lung 3; LUX‐Lung 6) (Table 1). Myelosuppression was consistently greater in the chemotherapy arms, while greater fatigue was seen in one comparison. Diarrhoea was worse with afatinib in both trials.

4. Quality of life

In LUX‐Lung 3, improvement was noted using the EORTC QLQ‐C30 scale in global health, physical, cognitive, and role function in favour of afatinib over cisplatin plus pemetrexed chemotherapy.

LUX‐Lung 6 also used the EORTC QLQ‐C30 scale and the lung cancer‐specific module QLQ‐LC13 with greater than 90% compliance. A greater percentage of participants showed improvement in global health scores/QoL scores (P < 0.0001), physical function (P < 0.0001), and social function (P < 0.0001) with afatinib when compared to cisplatin plus gemcitabine. Subgroup analysis showed delay in time to deterioration in cough, dyspnoea, and pain.

5. Symptom palliation

In the LUX‐Lung 3 trial, time‐to‐deterioration curves for cough and dyspnoea showed a significant effect in favour of afatinib (HR 0.60, 95% CI 0.41 to 0.87; P = 0.007) and (HR 0.68, 95% CI 0.50 to 0.93; P = 0.02), respectively. The HR for pain 0.83 (95% CI 0.62 to 1.10) was not statistically significant (P = 0.19).

In the LUX‐Lung 6 trial, time to deterioration for cough (HR 0.45; P = 0.0003), dyspnoea (HR 0.54; P < 0.0001), and pain (HR 0.70; P = 0.003) showed a significant effect in favour of afatinib (HR 0.56, 95% CI 0.41 to 0.77; P = 0.0002).

Cetuximab plus cytotoxic chemotherapy versus cytotoxic chemotherapy

Cetuximab plus paclitaxel or docetaxel plus carboplatin versus paclitaxel or docetaxel plus carboplatin: One trial considered this comparison (BMSO99).

Cetuximab plus vinorelbine plus cisplatin versus vinorelbine plus cisplatin: One trial considered this comparison (FLEX).

Primary outcome: Overall survival

We could not pool data for the two trials comparing cetuximab plus cytotoxic chemotherapy to cytotoxic chemotherapy, as one trial reported only an adjusted analysis (FLEX).

BMSO99 reported a HR of 1.62 (95% CI 0.54 to 4.84), indicating no significant difference in OS between the groups (Analysis 4.1).

FLEX reported a HR of 1.48 (95% CI 0.77 to 2.82), indicating no significant difference in OS between the groups (Analysis 4.1).

Secondary outcomes

1. Progression‐free survival

We could not pool data for the two trials comparing cetuximab plus cytotoxic chemotherapy to cytotoxic chemotherapy, as one trial reported only an adjusted analysis (FLEX).

BMSO99 reported a HR of 1.17 (95% CI 0.36 to 3.80), indicating no significant difference in PFS between the groups (Analysis 4.2).

FLEX reported a HR of 0.92 (95% CI 0.53 to 1.60), indicating no significant difference in PFS between the groups (Analysis 4.2).

2. Tumour response

The pooled treatment effect estimate (RR 1.43, 95% CI 0.83 to 2.47; I² = 40%; 2 trials) indicated no significant difference between the groups (Analysis 4.3).

3. Toxicity and adverse effects of treatment

The most commonly reported AEs in the cetuximab‐treated participants were neutropenia, leukopenia, febrile neutropenia, and fatigue (BMSO99; FLEX) (Table 1).

4. Quality of life

FLEX used the EORTC QLQ‐C30 and LCSS, and found no difference in QoL between the groups.

QoL was not available for the EGFR M+ subgroup in BMSO99.

5. Symptom palliation

Neither trial reported specifically on symptom palliation.

Toxicity and adverse effects of treatment ‐ general comments

The reporting of AEs differed across the 19 included trials. We described in Table 1 the trial‐defined reporting of AEs, and tabulated the three most frequently occurring grade 3 or 4 AE for both the intervention and comparator arms of each trial. The data reported were for overall trial populations, and therefore include non‐EGFR M+ participants in trials where these were unselected. The trials are grouped according to the EGFR‐targeted treatment employed (erlotinib, gefitinib, afatinib, cetuximab).

LUX‐Lung 3 and LUX‐Lung 6 reported three and two participants with interstitial lung disease, respectively (1%) in the afatinib arms.

The AEs associated with cytotoxic chemotherapy in all comparisons were neutropenia, fatigue, leukopenia, vomiting, anaemia, decreased appetite, diarrhoea, anorexia, thrombocytopenia, arthralgia, neuropathy, and dyspnoea.

Assessment of reporting biases

We have not included a funnel plot in the current review as we did not include a sufficient number of trials (n = 10) in any meta‐analysis. However, we devised and carried out a thorough search strategy to reduce the impact of publication bias.

Subgroup analyses

We did not include sufficient trials to allow subgroup analyses of smoking history, age, sex, ethnicity, type of mutation, or performance status.

Sensitivity analyses

We did not include sufficient trials in any one meta‐analysis to allow us to undertake the sensitivity analyses specified in the Methods section. However, where we detected moderate heterogeneity, we used a random‐effects model as a sensitivity analysis to compare results with the fixed‐effect model. We have reported these in the Effects of interventions section.

Network meta‐analysis

We considered that network meta‐analysis was not appropriate because of the different populations aross the included trials. We identified other barriers to conducting network meta‐analysis: two trials reported adjusted analyses (IPASS; NEJSG), whereas all other trials reported unadjusted analyses; participants in all trials were allowed to switch treatment after progression, and we had no information regarding how this was handled in the analysis for OS; and finally, the Kaplan‐Meier plots shown in the trial reports crossed in four trials, indicating that using a Cox proportional hazards model may not be appropriate.

Summary of findings table

We have presented tables for pooled analyses for the outcomes of OS and PFS: summary of findings Table for the main comparison; summary of findings Table 2; summary of findings Table 3.

Discussion

available in

Summary of main results

This review included 19 RCTs with a combined total of 2317 participants with EGFR M+ NSCLC. We identified four EGFR‐targeted treatments: erlotinib (eight trials); gefitinib (seven trials); afatinib (two trials); and cetuximab (two trials). We did not consider network meta‐analysis to be appropriate because of the different populations of included trials, the reporting of adjusted analyses versus unadjusted analyses, and the inappropriate use of the Cox proportional hazards model in some trials.

Our primary endpoint was overall survival (OS), and only one small (N = 97) trial reported a statistically significant OS gain (for participants treated with erlotinib plus cytotoxic chemotherapy versus cytotoxic chemotherapy alone) (FASTACT 2). None of the remaining 18 included trials demonstrated any OS benefit of targeted therapy compared with cytotoxic chemotherapy. No OS effect was demonstrated in pooled analyses of erlotinib in ENSURE, EURTAC, and OPTIMAL. Pooled analysis of two gefitinib trials, IPASS and NEJSG, and the two afatinib trials, LUX‐Lung 3 and LUX‐Lung 6, also showed no OS benefit. It is important to note that the majority of the included trials of anti‐EGFR monotherapy allowed participants to switch treatments on disease progression, which will have a confounding effect on any OS analysis.

For the secondary endpoint of progression‐free survival (PFS), a pooled analysis of four trials of erlotinib demonstrated a statistically significant benefit compared with cytotoxic chemotherapy (HR 0.30, 95% CI 0.24 to 0.38; 595 participants) (ENSURE; EURTAC; OPTIMAL; TORCH). Of the non‐pooled trials, for erlotinib versus cytotoxic chemotherapy, CHEN reported a non‐significant PFS effect of erlotinib (n = 24), and FASTACT 2 (n = 97) reported a significant PFS benefit for erlotinib (HR 0.25, 95% CI 0.16 to 0.39). The pooled analysis of gefitinib trials IPASS and NEJSG (N = 491) demonstrated a significant benefit of gefitinib compared with paclitaxel with carboplatin (HR 0.39, 95% CI 0.32 to 0.48). A single trial, WJTOG3405, also demonstrated a significant difference in PFS favouring gefitinib (HR 0.49, 95% CI 0.34 to 0.71). One other trial, First‐SIGNAL, demonstrated no statistically significant benefit of gefitinib compared with gemcitabine plus cisplatin (n = 42). The remaining two trials that featured gefitinib, INTACT 1 and INTACT 2, reported no difference between a regimen of gefitinib plus cytotoxic chemotherapy compared with cytotoxic chemotherapy plus placebo (n = 32). Heterogeneity was high in the pooled analyses of both erlotinib and gefitinib. Five trials showed a significant improvement in PFS for the tyrosine‐kinase inhibitor (TKI) in tumours harbouring the Del19 mutation compared to chemotherapy (EURTAC; IPASS; LUX‐Lung 3; NEJSG; OPTIMAL). We have not performed meta‐analysis of this mutation site‐specific data.

In the analysis of tumour response, a pooled analysis of 4 trials of erlotinib including 387 participants favoured treatment with erlotinib (RR 2.57, 95% CI 1.97 to 3.34) (EURTAC; GTOWG; OPTIMAL; TORCH). One trial of erlotinib plus cytotoxic chemotherapy (n = 97) also favoured treatment with erlotinib (FASTACT 2), whilst one other small trial of erlotinib versus cytotoxic chemotherapy reported no benefit of erlotinib (n = 24) (CHEN). For gefitinib, all 7 trials demonstrated a statistically significant benefit for gefitinib compared to cytotoxic chemotherapy: a pooled analysis of 4 trials including 648 participants yielded a RR of 1.87 (95% CI 1.60 to 2.19) (First‐SIGNAL; IPASS; NEJSG; WJTOG3405). Both afatinib trials (n = 709) reported a statistically significant benefit of afatinib compared with cytotoxic chemotherapy (LUX‐Lung 3; LUX‐Lung 6); the pooled analysis yielded a RR of 2.71 (95% CI 2.12 to 3.46). As for the PFS analyses, heterogeneity was high for the erlotinib and gefitinib pooled comparisons and low for the two afatinib trials. No benefit for cetuximab was reported for either trial (BMSO99; FLEX).

The most commonly reported adverse effects (AEs) for people treated with TKI monotherapy were rash, diarrhoea, paronychia, stomatitis/mucositis (afatinib), and rash, diarrhoea, and fatigue (erlotinib and gefitinib). These AEs are consistent with those listed in the Summary of Product Characteristics for these products, which include diarrhoea, rash, interstitial lung disease, liver impairment, and ocular disorders. Participants treated with cytotoxic chemotherapy experienced the AEs usually associated with this treatment, for example neutropenia, febrile neutropenia, leukopenia, and fatigue. However, it was difficult to accurately characterise and compare AEs across trials due to the different methods of reporting (definitions used and styles of reporting). This is particularly relevant to the rare but serious AE of interstitial lung disease. A recent meta‐analysis of erlotinib and gefitinib trials reported an incidence of 1.2% for interstitial lung disease with a mortality rate of 22.8% (Shi 2014). The data presented for afatinib suggest this complication occurs with equal frequency in all three TKIs, although no data on duration of therapy was provided. In addition, it should be noted that the AEs reported are relevant to an overall trial population, and in the 12 trials where EGFR M+ status was not an inclusion criterion, are drawn from a much larger population. However, our comparisons highlight the differences in the AEs associated with TKIs and cytotoxic chemotherapy (Pilkington 2012).

Six trials measured quality of life for participants with EGFR M+ tumours by a number of different methods (two comparing afatinib with cytotoxic chemotherapy, two comparing erlotinib with cytotoxic chemotherapy, and two comparing gefitinib with cytotoxic chemotherapy); all six trials reported a beneficial effect of the TKI compared to cytotoxic chemotherapy. All three TKIs showed symptom palliation of cough, pain, and dyspnoea, although the methodology used was not standardised.

Any benefit in survival has to be weighed against increased toxicity. The median number of chemotherapy cycles given in the control arms was four out of a planned six three‐weekly cycles. The oral agents were generally given until progression and appeared to be better tolerated. The median duration of therapy was estimated to be around 9 to 12 months. In the two gefitinib trials where data were presented, the number of participants discontinuing therapy was similar in the two groups, while in the EURTAC trial a higher proportion of participants on chemotherapy than on erlotinib discontinued due to toxicity.

Overall completeness and applicability of evidence

Median survival of people with advanced stages III, IV NSCLC is on the order of 12 months, and for adenocarcinomas 18 months. At present, there is no indication that increases in PFS fully translate into OS benefit, which is consistent with the evidence in the current literature base (Booth 2012). However, there was wide variation in the selection criteria for the included trials, including age, sex, smoking, and EGFR sequencing method. The later trials recruited participants only with proven EGFR mutations, and saw longer survival times. However, with the comparatively short survival in NSCLC, AEs and quality of life for either first‐line or second‐line treatments are important. The interpretation of OS was limited by cross‐over in most trials. From the limited data available on cross‐over at disease progression, the targeted agents and cytotoxics would appear to act on different cell populations.

Mutations in EGFR can be assessed by several methods including direct sequencing of the tumours, circulating tumour cells (Maheswaran 2008), or cell‐free DNA (Bai 2013). Firstly, heterogeneity in the proportion of malignant and normal/stromal cells in the tissues sampled may contribute to variation in the classification of tumours as EGFR M+ or EGFR wild type based on the location of the sample, as in the majority of trials in this review (Tsiatis 2010), and there is preliminary evidence of heterogeneity of mutation analysis with multiple tissue sampling (Bai 2013). Secondly, methodological issues in the assessment of EGFR mutations may contribute to false‐negative results (Vogelstein 2013). We excluded immunohistochemical‐only categorisation of mutation from this review.

Data on the types of mutations in relation to their sensitivity to targeted therapy is limited (EURTAC). Of the three common sites of mutation, there is evidence that tumours with codon 20 mutations are resistant to EGFR TKI, while tumours with exon 19 or L858R codon 21 mutations are sensitive to EGFR TKI (Yasuda 2011). The improved survival of exon 19 deletion patients with afatinib compared to cytotoxic chemotherapy suggests that further data will evolve based on more detailed molecular characterisation of EGFR M+ NSCLC (Yang 2014). The cetuximab trials assessed K‐RAS and HER‐2 mutations and demonstrated no predictive effect of the biomarkers (Linardou 2008). Non‐randomised trials have shown that some mutations, principally T790M in codon 20, may contribute to the development of acquired resistance to these agents (Kosaka 2006; Rosell 2011; Su 2012). Some trials did not include assessment of exons 18 and 20 mutations, although only four of the included trials excluded T790M mutations (FLEX; LUX‐Lung 3; LUX‐Lung 6; NEJSG).

With improving data on individualisation of treatment according to morphological and molecular criteria, patient choice may be a factor in the decision to accept significant toxicity (for example from cytotoxic chemotherapy) at an earlier or later stage of NSCLC management. This review provides strong data supporting first‐line EGFR TKI in people where EGFR mutation status is known to be positive. As mutation testing is not universally available, and the response time of reporting can be prolonged, chemotherapy may be an acceptable first‐line option when histological subtype and smoking history are known in patients with good performance status. Quality control of mutation profiling methodology and international agreement on standardisation would improve confidence in the use of EGFR TKIs in EGFR M+ patients.

There is some published evidence of ethnic differences in platinum‐based haematological toxicity, with Asian patients having a higher incidence of grade 3/4 neutropenia compared to non‐Asian patients, based on a pooled analysis of 11,271 participants in 50 phase II and III trials (Hasegawa 2011). It is less well established if there are ethnic differences in response to targeted therapies in the EGFR M+ subgroup, and there was wide variation in the ethnic composition of the reported trials. The majority of the data came from Asian patients, whose tumours may differ in genetic composition, both inherited and that acquired from carcinogen exposure, from non‐Asian patients.

Quality of the evidence

All the included trials were randomised, and the overall number of participants (n = 2317) in the 19 trials provides reasonable power to support the conclusions. The participants were spread across four different drug treatments (erlotinib, gefitinib, afatinib, and cetuximab), reducing the number providing data for each treatment.

We considered the quality of the evidence to be high for all comparisons (summary of findings Table for the main comparison; summary of findings Table 2; summary of findings Table 3). With the exception of FASTACT 2, all trials were of an open‐label design, however all but one trial, IPASS, reported independent review of radiographic outcomes.

The 'Risk of bias' table indicates a mixed risk of bias across the included trials for the majority of the assessment criteria, with most trials at unclear or high risk of bias (Figure 2; Figure 3). The two items considered to be at high risk of bias across the trials were related to blinding of treatment allocation for participants and personnel and blinding of outcome assessment. Blinding of participants and administrators is difficult to achieve in trials that compare oral therapy with intravenous chemotherapy treatments, and even if blinding procedures are implemented, the appearance of a rash (a common side effect of treatment with a TKI) would indicate the treatment regimen used. FASTACT 2 was blinded in both treatment allocation and imaging assessment. Blinding of outcome assessment is important when time‐to‐treatment‐failure outcomes, such as PFS, are the indicators of treatment efficacy, and blinded outcome assessment or blinded review of assessment should be part of the trial protocol. Of the large industry‐funded trials, OPTIMAL did not report blinding of outcome assessment for erlotinib, and neither did IPASS or WJTOG3405 for gefitinib. We acknowledge that some trials may have implemented such procedures but did not report them.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included trial.

The comparisons with cytotoxic chemotherapy were in general direct, but there was wide variation in the choice of cytotoxic chemotherapy in the comparator arm. This reflects variation in clinical practice, in particular performance status and comorbidity of the NSCLC populations. For example, single‐agent vinorelbine, used as the comparator in two of the smaller erlotinib trials (CHEN; GTOWG), is associated with lower toxicity than the more widely given doublet chemotherapy combinations used in the other trials, and participants in both CHEN and GTOWG were selected on the basis of age (older than 70) and not primarily performance status. The trials also varied in the extent to which they included never‐smokers or former smokers, and in the male/female ratio. The remaining major factor contributing to heterogeneity was ethnicity, as the eight trials recruiting exclusively in Asia contributed 64% of the participants. All of these factors may contribute to variation in drug handling of both cytotoxic chemotherapy and targeted therapy. Heterogeneity was high for assessment of PFS for erlotinib, gefitinib, and afatinib comparisons in the pooled data.

The results of this review should be interpreted cautiously. Just seven of the included trials recruited only people with EGFR mutations (n = 1672). This means that the data extracted from the remaining 12 trials (n = 645) are derived from subgroups, with all the issues that the interpretation of subgroup data entails. However, it is worth noting that the subgroup of EGFR M+ patients in the IPASS trial, at 261, was larger than the total trial population of four of the EGFR M+ only trials (EURTAC; NEJSG; OPTIMAL; WJTOG3405). It should be further noted that in four trials the tissue analyses were carried out retrospectively on a limited number of samples that were available at the end of the trial (BMSO99; FLEX; INTACT 1; INTACT 2). However, these four trials provided data from only 113 participants, and 80 of these were participants were from the cetuximab trials. We do not believe this factor has an impact on the overall conclusions with respect to the three TKIs.

The confidence limits of the PFS and OS plots were narrow, with the exception of the small trial of erlotinib (CHEN), and suggest the data are precise. We saw wider confidence limits for response, which may reflect the subjective nature of the assessment, even with external review, and current concerns PFS is the better endpoint for trial assessment where cross‐over is a factor (Booth 2012).

There is evidence that Asian patients have a different proportion of EGFR M+ and a differing relationship to smoking, which may imply there are differences in the biology of NSCLC between individuals of Asian and non‐Asian ethnicity . Of the 2317 participants reported on in this review, 1591 were recruited exclusively in trials conducted in Asian countries. We found no evidence that there is a different set of mutations in Asian and white patients, or differences in their toxicity profiles for the targeted or chemotherapy arms of the included trials.

Potential biases in the review process

We excluded trials that utilised EGFR‐targeted treatments but did not report any EGFR mutation testing of participants. However, inspection of review papers and reference lists indicated that in relation to four of these trials (BMSO99; FLEX; INTACT 1; INTACT 2), retrospective analyses of tissue samples from participants had taken place, the results of which were reported in papers separate to the original trial publication. It is possible that there are other retrospective analyses that we did not identify, however the patient population from any such analyses is likely to be small.

Agreements and disagreements with other studies or reviews

The results are in agreement with the meta‐analysis of Ku 2011, which compared gefitinib with first‐line chemotherapy. A more recent meta‐analysis of 14,570 participants given TKIs in first‐line, second‐line, and maintenance RCTs also supported gain in PFS in EGFR M+ participants treated with erlotinib and gefitinib (Lee 2013). This analysis included data on subgroups of participants (n = 67) from TALENT, TOPICAL, and TRIBUTE that were not available to us at the time of analysis. The Lee review analysed no data on participant characteristics, toxicity, and quality of life (Lee 2013). Their analysis combined the data from 10 first‐line trials in a meta‐analysis of OS and PFS, and showed an overall HR of 0.43 (95% CI 0.38 to 0.49; P < 0.001) for PFS and no effect on OS. As described above, we considered this pooling to be inappropriate on statistical grounds, as adjusted and unadjusted data were combined. An updated meta‐analysis by the same group focused on seven trials (ENSURE; EURTAC; LUX‐Lung 3; LUX‐Lung 6; NEJSG; OPTIMAL; WJTOG3405), and concluded that never‐smokers, those with tumours with exon 19 deletions, and women had a greater benefit from erlotinib than chemotherapy (Lee 2015). Other reviews have combined data from seven phase III trials, in Hasegawa 2015, and eight phase lll trials, in Haaland 2014, for first‐line chemotherapy, and confirmed the benefit in PFS and response. The data on benefit in non‐smokers is difficult to interpret in these studies. One network meta‐analysis of 12 trials combined first‐ and second‐line treatments, and concluded that erlotinib, gefitinib, and afatinib shared similar efficacy (Liang 2014). Our review of participants across 19 trials includes additional trials and comparable data from the 2317 EGFR M+ participants on afatinib, erlotinib, and gefitinib. A recent individual patient meta‐analysis of four RCTs of cetuximab, Pujol 2014, (including BMSO99 and FLEX) in NSCLC reported improved PFS in squamous cell cancers (based on a subgroup analysis) but not in non‐squamous carcinomas, although these data were not analysed by mutation status.

The prespecified analysis of the Del19 subgroup across a pooled analysis of both of the afatinib trials demonstrated an OS advantage for afatinib compared to chemotherapy in that subgroup, while the L858R subgroup (codon 21 mutation) showed no OS benefit (Yang 2014). Notably, cross‐over to afatinib in the control arm was not allowed, whilst in the majority of comparisons of erlotinib and gefitinib with cytotoxic chemotherapy, cross‐over to the corresponding TKI was permitted. Overall, there was a lack of data on OS benefit of EGFR inhibitors, but with a low confidence on this due to the inconsistency and imprecision of the results.

Figure 1

Study flow diagram.

Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included trial.

Analysis 1.1

Comparison 1 Erlotinib versus control, Outcome 1 Overall survival.

Analysis 1.2

Comparison 1 Erlotinib versus control, Outcome 2 Progression‐free survival.

Analysis 1.3

Comparison 1 Erlotinib versus control, Outcome 3 Tumour response.

Analysis 2.1

Comparison 2 Gefitinib versus CTX, Outcome 1 Overall survival.

Analysis 2.2

Comparison 2 Gefitinib versus CTX, Outcome 2 Progression‐free survival.

Analysis 2.3

Comparison 2 Gefitinib versus CTX, Outcome 3 Tumour response.

Analysis 3.1

Comparison 3 Afatinib versus CTX, Outcome 1 Overall survival.

Analysis 3.2

Comparison 3 Afatinib versus CTX, Outcome 2 Progression‐free survival.

Analysis 3.3

Comparison 3 Afatinib versus CTX, Outcome 3 Tumour response.

Analysis 4.1

Comparison 4 Cetuximab plus CTX versus CTX, Outcome 1 Overall survival.

Analysis 4.2

Comparison 4 Cetuximab plus CTX versus CTX, Outcome 2 Progression‐free survival.

Analysis 4.3

Comparison 4 Cetuximab plus CTX versus CTX, Outcome 3 Tumour response.

Analysis 5.1

Comparison 5 Gefitinib plus CTX versus CTX, Outcome 1 Progression‐free survival.

Summary of findings for the main comparison. Erlotinib vs control

First‐line treatment of advanced epidermal growth factor receptor (EGFR) mutation positive (M+) non‐squamous non‐small cell lung cancer (NSCLC): erlotinib comparisons
Patient or population: EGFR M+ patients with NSCLC Settings: oncology Intervention: erlotinib Comparison: control (cytotoxic chemotherapy)
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Control	Erlotinib
Overall survival	56 per 100	54 per 100 (46 to 63)	HR 0.95 (0.75, 1.22)	429 (3 studies)	High	All trials were open label but included blinded independent review
Progression‐free survival	73 per 100	33 per 100 (27 to 40)	HR 0.30 (0.24, 0.38)	595 (4 studies)	High	All trials were open label but included blinded independent review
The basis for the assumed risk is calculated as the event rate in the treatment group The corresponding risk is calculated as the assumed risk x the risk ratio (RR) of the intervention where RR = (1 ‐ exp(HR x ln(1 ‐ assumed risk)) )/assumed risk CI:* confidence interval; RR: risk ratio; HR: hazard ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.

Summary of findings for the main comparison. Erlotinib vs control

Summary of findings 2. Gefitinib vs paclitaxel + carboplatin

First‐line treatment of advanced epidermal growth factor receptor (EGFR) mutation positive (M+) non‐squamous non‐small cell lung cancer (NSCLC): gefitinib comparisons
Patient or population: EGFR M+ patients with NSCLC Settings: oncology Intervention: gefitinib Comparison: paclitaxel + carboplatin
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Paclitaxel + carboplatin	Gefitinib
Overall survival	67 per 100	66 per 100 (58 to 73)	HR 0.95 (0.77 to 1.18)	489 (2 studies)	High	Both trials were open label. IPASS did not report independent blinded review
Progression‐free survival	89 per 100	57 per 100 (50 to 65)	HR 0.39 (0.32 to 0.48)	485 (2 studies)	High	Both trials were open label. IPASS did not report independent blinded review
The basis for the assumed risk is calculated as the event rate in the treatment group The corresponding risk is calculated as the assumed risk x the risk ratio (RR) of the intervention where RR = (1 ‐ exp(HR x ln(1 ‐ assumed risk)) )/assumed risk CI:* confidence interval; RR: risk ratio; HR: hazard ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.

Summary of findings 2. Gefitinib vs paclitaxel + carboplatin

Summary of findings 3. Afatinib vs chemotherapy

First‐line treatment of advanced epidermal growth factor receptor (EGFR) mutation positive (M+) non‐squamous non‐small cell lung cancer (NSCLC): afatinib comparisons
Patient or population: EGFR M+ patients with NSCLC Settings: oncology Intervention: afatinib Comparison: cytotoxic chemotherapy
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Cytotoxic chemotherapy	Afatinib
Overall survival	46 per 100	44 per 100 (37 to 52)	HR 0.93 (0.74 to 1.17)	709 (2 studies)	High	Both trials were open label but included blinded independent central review
Progression‐free survival	56 per 100	29 per 100 (24 to 35)	HR 0.42 (0.34 to 0.53)	709 (2 studies)	High	Both trials were open label but included blinded independent central review
The basis for the assumed risk is calculated as the event rate in the treatment group The corresponding risk is calculated as the assumed risk x the risk ratio (RR) of the intervention where RR = (1 ‐ exp(HR x ln(1 ‐ assumed risk)) )/assumed risk CI:* confidence interval; RR: risk ratio; HR: hazard ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.

Summary of findings 3. Afatinib vs chemotherapy

Table 1. Adverse events ‐ most commonly occurring grade 3 & 4

Study

Definition of AE

Population

Top AE (listed according to intervention)

Second top AE (listed according to intervention)

Third top AE (listed according to intervention)

Top 3 AEs (listed according to comparator)

Afatinib trials

LUX‐Lung 3

Grade >= 3 CTC (V3)

AEs that were reported in > 10% of participants in either group and if there was a >= 10% difference between the groups

EGFR M+ only

Rash/acne:

16.2% (AFA) vs 0% (cytotoxic chemotherapy)

Diarrhoea:

14.4% (AFA) vs 0% (cytotoxic chemotherapy)

Paronychia:

11.4% (AFA) vs 0% (cytotoxic chemotherapy)

Neutropenia: 18% vs 0.4%

Fatigue: 12.6% vs 1.3%

Leukopenia: 8.1% vs 0.4%

LUX‐Lung 6

CTC (V3)

Events are included if reported for >= 1% of participants in any treatment group

EGFR M+ only

Rash/acne:

14.6% (AFA) vs 0% (cytotoxic chemotherapy)

Diarrhoea:

5.4% (AFA) vs 0% (cytotoxic chemotherapy)

Stomatitis/mucositis:

5.4% (AFA) vs 0% (cytotoxic chemotherapy)

Neutropenia: 26.5% vs 0.4%

Vomiting: 19.4% vs 0.8%

Leukopenia: 15.1% vs 0.4%

Erlotinib trials

CHEN

Incidence rate >= 10%

Unselected population

Rash:

64.9% (ERL) vs NR (cytotoxic chemotherapy)

Diarrhoea:

29.8% (ERL) vs NR (cytotoxic chemotherapy)

Mouth ulceration:

14% (ERL) vs NR (cytotoxic chemotherapy)

Anorexia: 26.3% vs NR

Diarrhoea: 12.3% vs NR

Vomiting: 10.5% vs NR

ENSURE

Grade ≥ 3

≥ 5% in either arm

EGFR M+ only

Rash:

6.4% (ERL) vs 1% (cytotoxic chemotherapy)

Neutropenia, leukopenia,

anaemia:

All 0.9% (ERL) vs 25%, 14.4%, 12.5% respectively (cytotoxic chemotherapy)

‐

Neutropenia: 25% vs 0.9%

Leukopenia: 14.4% vs 0.9%

Anaemia: 12.5% vs 0.9%

EURTAC

Grade 3/4 CTC (V3)

Common AEs

EGFR M+ only

Rash:

13% (ERL) vs 0% (cytotoxic chemotherapy)

Fatigue:

6% (ERL) vs 20% (cytotoxic chemotherapy)

Diarrhoea:

5% (ERL) vs 0% (cytotoxic chemotherapy)

Neutropenia: 22% vs 0%

Fatigue: 20% vs 6%

Thrombocytopenia: 14% vs 0%

FASTACT 2

Grade 3/4 CTC (V3)

Most commonly reported

Unselected population

Neutropenia:

29% (ERL) vs 25% (cytotoxic chemotherapy)

Thrombocytopenia

14% (ERL) vs 14% (cytotoxic chemotherapy)

Anaemia:

11% (ERL) vs 9% (cytotoxic chemotherapy)

Neutropenia: 25% vs 29%

Thrombocytopenia: 14% vs 14%

Anaemia: 9% vs 11%

GTOWG

Grade 3/4

Unselected population

Rash:

12% (ERL) vs 0% (cytotoxic chemotherapy)

Diarrhoea:

6% (ERL) vs 2% (cytotoxic chemotherapy)

Constitutional symptoms:

3% (ERL) vs 5% (cytotoxic chemotherapy)

Neutropenia: 36% vs 0%

Leukocytes: 33% vs 0%

Haemoglobin: 11% vs 0.7%

OPTIMAL

Grade 3/4 CTC (V3)

AEs occurred in 3% or more in either treatment group

EGFR M+ only

Increased ALT:

4% (ERL) vs 1% (cytotoxic chemotherapy)

Skin rash:

2% (ERL) vs 0% (cytotoxic chemotherapy)

Diarrhoea:

1% (ERL) vs 0% (cytotoxic chemotherapy)

Neutropenia: 42% vs 0%

Thrombocytopenia: 40% vs 0%

Anaemia: 13% vs 0%

TOPICAL

CTC (V3)

Specific AEs grade 3 or 4

Unselected population

Dyspnoea:

59% (ERL) vs 64% (PLA)

Fatigue:

23% (ERL) vs 23% (PLA)

Diarrhoea:

8% (ERL) vs 1% (cytotoxic chemotherapy)

Dyspnoea:

64% vs 59%

Fatigue:

23% vs 23%

Anorexia: 5% vs 5%

TORCH

Worst toxicity experienced with first‐line treatment alone

Unselected population

Skin rash:

11% (ERL) vs 0% (cytotoxic chemotherapy)

Pulmonary toxicity:

9% (ERL) vs 6% (cytotoxic chemotherapy)

Fatigue:

8% (ERL) vs 12% (cytotoxic chemotherapy)

Neutropenia: 21% vs 0%

Thrombocytopenia: 12% vs 0%

Fatigue: 12% vs 8%

Gefitinib trials

First‐SIGNAL

Grade 3 or 4 CTC (V3)

Unselected

population

Rash:

29.3% (GEF) vs 2% (cytotoxic chemotherapy)

Anorexia:

13.8% (GEF) vs 57.3% (cytotoxic chemotherapy)

AST:

11.3% (GEF) vs 2% (cytotoxic chemotherapy)

Anorexia: 57.3% vs 13.9%

Neutropenia: 54% vs 1.9%

Fatigue: 45.3% vs 10.1%

INTACT 1

Grade 3/4 CTC

Commonly occurring AEs

Unselected

population

Thrombocytopenia*:

5.8% (GEF + cytotoxic chemotherapy) vs 5.6% (cytotoxic chemotherapy)

Rash:

3.6% (GEF + cytotoxic chemotherapy) vs 1.1% (cytotoxic chemotherapy)

Diarrhoea:

3.6% (GEF + cytotoxic chemotherapy) vs 2.3% (cytotoxic chemotherapy)

Thrombocytopenia*: 5.6% vs 5.8%

Leukopenia: 2.5% vs 3.3%

Diarrhoea: 2.3% vs 3.6%

INTACT 2

Grade 3/4 CTC (V2)

Common drug‐related AEs

Unselected

population

Diarrhoea:

9.9% (GEF + cytotoxic chemotherapy) vs 2.9% (cytotoxic chemotherapy)

Neutropenia:

6.7% (GEF + cytotoxic chemotherapy) vs 5.9% (cytotoxic chemotherapy)

Rash:

3.2% (GEF + cytotoxic chemotherapy) vs 1.5% (cytotoxic chemotherapy)

Neutropenia: 5.9% vs 6.7%

Diarrhoea: 2.9% vs 9.9%

Vomiting: 2.3% vs 2%

IPASS

Grade 3, 4, or 5 CTC (V3)

At least 10% of participants in either treatment group and at least a 5% difference between arms

Unselected

population

Diarrhoea:

3.8% (GEF) vs 1.4% (cytotoxic chemotherapy)

Any neutropenia:

3.7% (GEF) vs 67.1% (cytotoxic chemotherapy)

Rash:

3.1% (GEF) vs 0.8% (cytotoxic chemotherapy)

Any neutropenia: 67.1% vs 3.7%

Leukopenia: 35% vs 1.5%

Anaemia: 10.6% vs 2.2%

NEJSG

Grade >= 3 CTC (V3)

At least 10% of participants in either treatment group and at least a 5% difference between arms

EGFR M+ only

ATE:

26.3% (GEF) vs 0.9% (cytotoxic chemotherapy)

Rash:

5.3% (GEF) vs 2.7% (cytotoxic chemotherapy)

Appetite loss:

5.3% (GEF) vs 6.2% (cytotoxic chemotherapy)

Neutropenia: 65.5% vs 0.9%

Arthralgia: 7.1% vs 0.9%

Neuropathy: 6.2% vs 0%

Appetite loss: 6.2% vs 5.3%

WJTOG3405

Grade >= 3 CTC (V3)

AEs occurred in 10% of either of the treatment groups

EGFR M+ only

ALT/AST:

27.5% (GEF) vs 2.3% (cytotoxic chemotherapy)

Rash:

2.3% (GEF) vs 0% (cytotoxic chemotherapy)

Fatigue:

2.3% (GEF) vs 2.3% (cytotoxic chemotherapy)

Neutropenia: 84% vs 0%

Leucocytopenia: 50% vs 0%

Anaemia: 17% vs 0%

Yu 2014

Grade 3+

Participants with at least 1 AE

Unselected

population

Rash:

16% (GEF + cytotoxic chemotherapy) vs 0% (cytotoxic chemotherapy)

Vomiting:

10% (GEF) vs 8% (cytotoxic chemotherapy)

Neutropenia:

10% (GEF) vs 12% (cytotoxic chemotherapy)

Neutropenia: 12% vs 10%

Nausea: 8% vs 5%

Vomiting: 8% vs 10%

Cetuximab trials

BMSO99

Grade 3/4 CTC (V3)

Most frequent and relevant grade 3/4 AEs

Unselected population

Neutropenia:

62.5% (CET + cytotoxic chemotherapy) vs 56% (cytotoxic chemotherapy)

Leukopenia:

43.8% (CET + cytotoxic chemotherapy) vs 30.7% (cytotoxic chemotherapy)

Fatigue:

15.1% (CET + cytotoxic chemotherapy) vs 12.2% (cytotoxic chemotherapy)

Same AEs as intervention

FLEX

Grade 3/4 CTC (V2)

AEs that were reported in > 5% of participants (G3/G4) or > 1% (G4) or AEs of special interest in either group

EGFR M+ expressing

Neutropenia:

53% (CET + cytotoxic chemotherapy) vs 51% (cytotoxic chemotherapy)

Leukopenia:

25% (CET + cytotoxic chemotherapy) vs 19% (cytotoxic chemotherapy)

Febrile neutropenia:

22% (CET + cytotoxic chemotherapy) vs 15% (cytotoxic chemotherapy)

Neutropenia: 52% (cytotoxic chemotherapy) vs 52% CET + cytotoxic chemotherapy

Leukopenia: 19% (cytotoxic chemotherapy) vs 25% (CET vs cytotoxic chemotherapy)

Anaemia: 16% (cytotoxic chemotherapy) vs 1% (CET + cytotoxic chemotherapy)

PLA: placebo

*Neutropenia was also reported as 5.8% for G3/4; as this rate was higher than the rate for all participants (5%) it was not included in the table.

Table 1. Adverse events ‐ most commonly occurring grade 3 & 4

Comparison 1. Erlotinib versus control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Overall survival Show forest plot	5		Hazard Ratio (Random, 95% CI)	Subtotals only

1.1 Erlotinib versus CTX	3		Hazard Ratio (Random, 95% CI)	0.95 [0.75, 1.22]
1.2 Erlotinib versus vinorelbine	1		Hazard Ratio (Random, 95% CI)	2.16 [0.58, 8.10]
1.3 Erlotinib plus CTX versus CTX plus placebo	1		Hazard Ratio (Random, 95% CI)	0.48 [0.27, 0.85]
2 Progression‐free survival Show forest plot	6		Hazard Ratio (Fixed, 95% CI)	Subtotals only

2.1 Erlotinib versus CTX	4		Hazard Ratio (Fixed, 95% CI)	0.30 [0.24, 0.38]
2.2 Erlotinib versus vinorelbine	1		Hazard Ratio (Fixed, 95% CI)	0.55 [0.21, 1.46]
2.3 Erlotinib plus CTX versus CTX plus placebo	1		Hazard Ratio (Fixed, 95% CI)	0.25 [0.16, 0.39]
3 Tumour response Show forest plot	7		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only

3.1 Erlotinib versus CTX	5	593	Risk Ratio (M‐H, Fixed, 95% CI)	2.26 [1.85, 2.76]
3.2 Erlotinib versus vinorelbine	1	24	Risk Ratio (M‐H, Fixed, 95% CI)	0.83 [0.19, 3.67]
3.3 Erlotinib versus erlotinib plus CTX	0	0	Risk Ratio (M‐H, Fixed, 95% CI)	0.0 [0.0, 0.0]
3.4 Erlotinib plus CTX versus CTX plus placebo	1	97	Risk Ratio (M‐H, Fixed, 95% CI)	5.74 [2.86, 11.50]

Comparison 1. Erlotinib versus control

Comparison 2. Gefitinib versus CTX

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Overall survival Show forest plot	4		Hazard Ratio (Fixed, 95% CI)	Subtotals only

1.1 Gefitinib versus gemcitabine plus cisplatin	1		Hazard Ratio (Fixed, 95% CI)	1.04 [0.50, 2.20]
1.2 Gefitinib versus paclitaxel plus carboplatin	2		Hazard Ratio (Fixed, 95% CI)	0.95 [0.77, 1.18]
1.3 Gefitinib versus docetaxel plus cisplatin	1		Hazard Ratio (Fixed, 95% CI)	1.25 [0.88, 1.78]
2 Progression‐free survival Show forest plot	4		Hazard Ratio (Fixed, 95% CI)	Subtotals only

2.1 Gefitinib versus gemcitabine plus cisplatin	1		Hazard Ratio (Fixed, 95% CI)	0.54 [0.27, 1.10]
2.2 Gefitinib versus paclitaxel plus carboplatin	2		Hazard Ratio (Fixed, 95% CI)	0.39 [0.32, 0.48]
2.3 Gefitinib versus docetaxel plus cisplatin	1		Hazard Ratio (Fixed, 95% CI)	0.49 [0.34, 0.71]
3 Tumour response Show forest plot	4	648	Risk Ratio (M‐H, Fixed, 95% CI)	1.87 [1.60, 2.19]

3.1 Gefitinib versus gemcitabine plus cisplatin	1	42	Risk Ratio (M‐H, Fixed, 95% CI)	2.26 [1.17, 4.34]
3.2 Gefitinib versus paclitaxel plus carboplatin	2	489	Risk Ratio (M‐H, Fixed, 95% CI)	1.83 [1.54, 2.18]
3.3 Gefitinib versus docetaxel plus cisplatin	1	117	Risk Ratio (M‐H, Fixed, 95% CI)	1.93 [1.26, 2.94]

Comparison 2. Gefitinib versus CTX

Comparison 3. Afatinib versus CTX

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Overall survival Show forest plot	2		Hazard Ratio (Fixed, 95% CI)	0.93 [0.74, 1.17]

1.1 Afatinib versus pemetrexed plus cisplatin	1		Hazard Ratio (Fixed, 95% CI)	0.91 [0.66, 1.25]
1.2 Afatinib versus gemcitabine plus cisplatin	1		Hazard Ratio (Fixed, 95% CI)	0.95 [0.68, 1.33]
2 Progression‐free survival Show forest plot	2		Hazard Ratio (Fixed, 95% CI)	0.42 [0.34, 0.53]

2.1 Afatinib versus pemetrexed plus cisplatin	1		Hazard Ratio (Fixed, 95% CI)	0.58 [0.43, 0.78]
2.2 Afatinib versus gemcitabine plus cisplatin	1		Hazard Ratio (Fixed, 95% CI)	0.28 [0.20, 0.39]
3 Tumour response Show forest plot	2	709	Risk Ratio (M‐H, Fixed, 95% CI)	2.71 [2.12, 3.46]

3.1 Afatinib versus pemetrexed plus cisplatin	1	345	Risk Ratio (M‐H, Fixed, 95% CI)	2.48 [1.74, 3.54]
3.2 Afatinib versus gemcitabine plus cisplatin	1	364	Risk Ratio (M‐H, Fixed, 95% CI)	2.92 [2.08, 4.09]

Comparison 3. Afatinib versus CTX

Comparison 4. Cetuximab plus CTX versus CTX

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Overall survival Show forest plot	2		Hazard Ratio (Fixed, 95% CI)	Subtotals only

1.1 Cetuximab plus paclitaxel or docetaxel plus carboplatin versus paclitaxel or docetaxel plus carboplatin	1		Hazard Ratio (Fixed, 95% CI)	1.62 [0.54, 4.84]
1.2 Cetuximab plus vinorelbine plus cisplatin versus vinorelbine plus cisplatin	1		Hazard Ratio (Fixed, 95% CI)	1.48 [0.77, 2.82]
2 Progression‐free survival Show forest plot	2		Hazard Ratio (Fixed, 95% CI)	Subtotals only

2.1 Cetuximab plus paclitaxel or docetaxel plus carboplatin versus paclitaxel or docetaxel plus carboplatin	1		Hazard Ratio (Fixed, 95% CI)	1.17 [0.36, 3.80]
2.2 Cetuximab plus vinorelbine plus cisplatin versus vinorelbine plus cisplatin	1		Hazard Ratio (Fixed, 95% CI)	0.92 [0.53, 1.60]
3 Tumour response Show forest plot	2	81	Risk Ratio (M‐H, Fixed, 95% CI)	1.43 [0.83, 2.47]

3.1 Cetuximab plus paclitaxel or docetaxel plus carboplatin versus paclitaxel or docetaxel plus carboplatin	1	17	Risk Ratio (M‐H, Fixed, 95% CI)	4.5 [0.63, 32.38]
3.2 Cetuximab plus vinorelbine plus cisplatin versus vinorelbine plus cisplatin	1	64	Risk Ratio (M‐H, Fixed, 95% CI)	1.19 [0.67, 2.11]

Comparison 4. Cetuximab plus CTX versus CTX

Comparison 5. Gefitinib plus CTX versus CTX

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Progression‐free survival Show forest plot	1		Hazard Ratio (Fixed, 95% CI)	0.20 [0.05, 0.75]

Comparison 5. Gefitinib plus CTX versus CTX