Pooled analysis of clinical outcome for EGFR TKI-treated patients with EGFR mutation-positive NSCLC

Patients with non-small-cell lung cancer (NSCLC) appear to gain particular benefit from treatment with epidermal growth factor receptor (EGFR) tyrosine-kinase inhibitors (TKI) if their disease tests positive for EGFR activating mutations. Recently, several large, controlled, phase III studies have been published in NSCLC patients with EGFR mutation-positive tumours. Given the increased patient dataset now available, a comprehensive literature search for EGFR TKIs or chemotherapy in EGFR mutation-positive NSCLC was undertaken to update the results of a previously published pooled analysis. Pooling eligible progression-free survival (PFS) data from 27 erlotinib studies (n = 731), 54 gefitinib studies (n = 1802) and 20 chemotherapy studies (n = 984) provided median PFS values for each treatment. The pooled median PFS was: 12.4 months (95% accuracy intervals [AI] 11.6–13.4) for erlotinib-treated patients; 9.4 months (95% AI 9.0–9.8) for gefitinib-treated patients; and 5.6 months (95% AI 5.3–6.0) for chemotherapy. Both erlotinib and gefitinib resulted in significantly longer PFS than chemotherapy (permutation testing; P = 0.000 and P = 0.000, respectively). Data on more recent TKIs (afatinib, dacomitinib and icotinib) were insufficient at this time-point to carry out a pooled PFS analysis on these compounds. The results of this updated pooled analysis suggest a substantial clear PFS benefit of treating patients with EGFR mutation-positive NSCLC with erlotinib or gefitinib compared with chemotherapy.


Introduction
The identification of new molecular targets for the treatment of lung cancer has revolutionized treatment paradigms for this disease. Non-small-cell lung cancer (NSCLC) therapy has benefited from the discovery of the epidermal growth factor receptor (EGFR) as a key mediator of cell proliferation. Efforts to identify agents to target EGFR led to the development of the EGFR tyrosine-kinase inhibitors (TKIs), which target the tyrosine-kinase (TK) domain of the receptor. Two EGFR TKIs have been approved for use in NSCLC in Europe and North America, erlotinib and gefitinib. Erlotinib has shown efficacy for second-or third-line treatment of NSCLC [1], as maintenance therapy [2] and for the first-line treatment of EGFR mutation-positive disease [3,4]. Gefitinib has shown efficacy for the treatment of locally advanced or metastatic NSCLC with activating EGFR mutations [5][6][7].
Low mutation rate and low availability of tumour samples limited the sample size for most of the efficacy analyses of erlotinib or gefitinib in patients with EGFR mutation-positive tumours. This prompted a retrospective pooled analysis by Paz-Ares et al. in 2010 [8], which reinforced the evidence that the preferential use of EGFR TKIs in patients with EGFR mutation-positive tumours may be warranted. Several additional large, phase III studies have since reported data in EGFR mutation-positive populations, including the WJTOG/802 [9], NEJ002 [10], OPTIMAL [3] and EURTAC [4] studies. Other anti-EGFR agents are also under investigation. Afatinib (an irreversible HERfamily blocker), dacomitinib (an irreversible TKI of EGFR, HER2 and HER4) and icotinib (an EGFR TKI) have shown activity in EGFR mutation-positive NSCLC [11][12][13][14][15][16]. To date, the datasets for these compounds remain relatively limited although the results of one phase III trial have been recently reported [11].
The molecular biology of the EGFR mutation was reviewed extensively in the previous pooled analysis publication [8]. Briefly, EGFR plays a role in the mediation of cell signalling by regulating proliferation, angiogenesis and apoptosis [17,18]. Ninety per cent of NSCLC EGFR mutations comprise a leucine to arginine substitution at position 858 in exon 21 (L858R) or various deletion mutations in exon 19 [19][20][21][22][23]. Epidermal growth factor receptor mutations alter the TK pocket of the receptor, enhancing its sensitivity to EGFR TKIs.
Epidermal growth factor receptor mutations are found in around 10% of NSCLC in Caucasians and 30% of NSCLC in East-Asians [22]. Correlations between EGFR mutation-positive status and clinical characteristics have been reported, however, with the mutations being more common in the tumours of never-smokers and females, and in adenocarcinomas [22,24,25]. This correlation is not exclusive and patients cannot be assumed to have EGFR mutation-positive disease based on clinical profile alone. Therefore, EGFR mutation testing is essential; currently the European Society for Medical Oncology guidelines indicate that EGFR mutation testing is recommended as standard in non-squamous NSCLC [26]. Tumour specimens from curative surgery or bronchial biopsy are the gold standard for testing, but less than one-third of patients are suitable for surgery [27] and bronchial biopsy is impractical for poorly accessible tumours. Cytological samples, such as fine-needle aspirates, bronchial brushings, serum, plasma, circulating tumour cells and pleural effusion samples have all been used for EGFR mutation testing, but are considered less reliable because of heterogeneity of tissue samples and sparse cellularity [28].
Genotyping of EGFR mutations can be accomplished by several techniques. Direct DNA sequencing e.g. pyrosequencing and dideoxy 'Sanger' sequencing, will reveal any mutation. Detection by PCR (e.g. PCR-fragment length analysis) or real-time quantitative PCR (qPCR) is routinely employed to detect known, pre-specified EGFR mutations [29]. Locked nucleic acid genotyping is also used. In the clinical setting, rapid diagnostic testing may be employed with real-time PCR kits, which detect a specific number of mutations. Sequencing is still required to detect the rarer mutations. This pooled analysis focuses primarily on studies that included patients with exon 19 or exon 21 mutated NSCLC; multiple techniques have proven efficacious at detecting these classical mutations with high specificity and variable levels of detection. Cases identified by Sanger sequencing or highly sensitive methods appear to respond similarly to EGFR TKI [3,4].
The increased number of studies that have examined the efficacy of the EGFR TKIs in patients with exon 19 or exon 21 mutated NSCLC provides an expanded dataset for analysis. This paper describes an updated literature search for clinical studies of erlotinib, gefitinib and other EGFR TKIs in patients with EGFR mutation-positive NSCLC, and reports the results of a pooled analysis of erlotinib, gefitinib and chemotherapy, with the aim of providing updated median pooled progression-free survival (PFS) values. This study should help to provide robust recommendations for the clinical management of patients in this important patient subset.

Selection criteria
All prospective and retrospective studies that examined erlotinib, gefitinib, icotinib, afatinib or dacomitinib as single-agent therapy or chemotherapy as single-or multiple-agent treatment for patients with EGFR mutationpositive NSCLC were eligible for inclusion. The studies were not critically assessed for methodology of EGFR mutation status determination.

Literature search strategy
The literature was reviewed to identify studies for inclusion in the pooled analysis.  [8] were checked for updates (e.g. updated analyses or full papers). Only English language papers from 2004 or later were included. The papers were initially filtered by manually scanning the titles. Abstracts of the remaining papers were reviewed and filtered further. The remaining papers were then reviewed in full. The search excluded studies where: the results were only given graphically; two TKIs were given in sequence; PFS data for EGFR TKIs were presented as a class and not split by drug; patients were treated in the maintenance or adjuvant setting; EGFR TKI or chemotherapy was used in combination with any other therapy (including surgery); the PFS/time-to-progression (TTP) was not given for EGFR mutation-positive NSCLC; information was ambiguous; only updates were reported; EGFR mutation status analysis was performed with blood samples; results were presented for patients only with 'other' EGFR mutations (i.e. non-exon 19 deletion or L858R mutation); or patients received a non-standard dose. The search also excluded studies where patients were selected for EGFR mutation status plus: another biomarker; specific types or patterns of metastases; having very poor performance status (3+); having benefited from an EGFR TKI previously (including long-term responders). Individual case studies were excluded.

Statistical analysis
The methodology was as described in Paz-Ares et al. [8], and is reported only briefly here. Individual data points were not available for PFS/TTP, therefore calculation of a pooled median PFS was made from the study medians. In some studies, a median PFS was not available, and simplifying assumptions (exponential distribution) were made to approximate the study median. In one study, a mean PFS value instead of median PFS was reported; based on the simplifying assumption of an underlying exponential distribution the median is simply the mean multiplied by ln (2). In some studies PFS was reported as 'after T(time) months a fraction of the patients were without progression. . ..'. The median PFS was estimated based on the assumptions of an exponential distribution by T/{ln(PFS) ln(x)}. Finally, in some cases multiple reports of PFS by time and percentage were reported. Here, the average of the multiple medians was taken to approximate the overall median in this study. The pooled median PFS (MPFS all ) was then obtained by a weighted average of the single study medians, which was calculated by multiplying the study median (MPFS (i) ) by the size of the study and summing over all studies. The result was divided by the total number of patients in all the studies (N all ): Valid confidence intervals (CIs) to assess inherent variation can only be calculated when individual data are given. These were not available for all studies so a surrogate 'accuracy interval' was calculated to reflect the comparative accuracy of the median estimates. The reported medians were treated as maximum likelihood estimates of the parameters of the exponential distribution to determine the pertaining confidence bands as 'accuracy intervals'; the 90% and 95% confidence bands were used as surrogate 'accuracy intervals' (single studies and pooled median estimates, respectively).
Random permutations [30] across studies were generated for 20,000 runs to test the null hypothesis that there is no difference between treatments. This comparative test is statistically valid, but only refers to the given study pool (conditional test) and cannot be readily extrapolated to the total patient population.
The potential for publication bias was assessed by using funnel plots with the pertaining accuracy intervals.

Sensitivity analysis
Sensitivity analyses were performed to assess the adequacy of the calculated accuracy intervals by using a resampling technique ('bootstrap') (e.g. Hesterberg et al. [30]) with 5000 runs.

Breakdown of eligible studies
The number of studies identified and excluded is shown in Figure 1. In total, 92 papers or abstracts contained PFS values that were eligible for this pooled analysis. There were 26 studies that evaluated erlotinib, 54 studies that evaluated gefitinib, 20 that evaluated chemotherapy, two that evaluated icotinib and one that evaluated afatinib (Table 1). Of the 27 erlotinib studies, 10 were first line, 17 of 54 gefitinib studies were first line and 17 of 20 chemotherapy studies were first line. There were two erlotinib phase III trials, seven gefitinib phase III trials, nine chemotherapy phase III trials and one icotinib phase III trial. There were 10 retrospective erlotinib trials (n = 127), 26 retrospective gefitinib trials (n = 861), and 11 retrospective chemotherapy trials (n = 439). The total number of patients included in the pooled analysis was 3521 (731 were treated with erlotinib, 1802 were treated with gefitinib and 984 were treated with chemotherapy). Afatinib and icotinib were not included in the pooled analysis calculations, but studies were identified that included 129 afatinib-treated patients (US and Taiwanese) and 29 icotinib-treated patients (all Chinese). In studies where mutation types were reported individually the most common EGFR mutations were exon 19 deletions (53%) and L858R mutations (38%).
There was a mixture of ethnicities included in the pooled analysis. Since only patients with EGFR activating mutations were included, no effect of ethnicity on efficacy was expected. Subgroup analyses revealed no striking differences between Asian and Caucasian patients with EGFR activating mutations and therefore, no adjustments were made in the analysis for ethnicity.
When the median PFS data are examined individually, there is a trend for erlotinib and gefitinib to report longer PFS times than chemotherapy (Fig. 2). This trend is confirmed when the pooled median PFS values are considered (Fig. 3). When analysed for significance by permutation testing (Table 2) there was a statistically significant increase in PFS for erlotinib compared with chemotherapy in the first line (P = 0.000), in lines other than first (P = 0.0022) and in all lines (P = 0.000). There was also a statistically significant increase in PFS for gefitinib compared with chemotherapy in the first line (P = 0.000), in lines other than first (P = 0.0039) and in all lines (P = 0.000). There were only three chemotherapy studies (n = 116) in treatment lines other than first; which limits the interpretation of this result, despite the significant P-value. However, the pooled median PFS value of 4.1 months for these three studies was not     (Table 3), the pooled median PFS for chemotherapy was different when given as predominantly firstline treatment versus other lines of treatment (5.8 versus 4.1 months, respectively, P = 0.012). The analysis did not discriminate between single-agent chemotherapy and doublet chemotherapy. The pooled median PFS values for erlotinib and gefitinib were not statistically different between lines of treatment (erlotinib: 12.0 versus 12.9 months, for first and other lines, respectively, P = 0.678; gefitinib: 9.7 versus 9.1 months, for first and other lines, respectively, P = 0.283).
Statistical analysis of a pooled median PFS could not be established for afatinib and icotinib because of lack of data. In one study for afatinib (n = 129), the PFS was 14 months [110], and in two studies (n = 27) and (n = 7) the PFS with icotinib was 6.5 months and 4.6 months, respectively [15,16] (Table 1). No eligible dacomitinib studies were identified at the time of analysis.
Publication bias was assessed by using funnel plots with PFS/TTP as the outcome. These were symmetrical for each of the treatment groups ( Fig. 4A-C).

Discussion
The dataset analyzed here was almost double the size of that previously assessed [8]. The patient number was updated from 365 to 731 in the erlotinib arm, from 1069 to 1802 in the gefitinib arm and from 375 to 984 in the chemotherapy studies. Progression-free survival was again chosen as the end-point to assess. Because of high levels of crossover in post-study therapy, the use of overall survival as an end-point was not considered to be able to discern differences between treatments. This analysis indicates that PFS is longer in patients with EGFR mutation-positive NSCLC when treated with erlotinib (12.4 months) or gefitinib (9.4 months), compared with conventional chemotherapy (5.6 months). Permutation testing indicated that the difference in PFS was statistically significant, but this should be interpreted carefully, given that this significance applies to the current study pool and it cannot be readily extrapolated to the total patient population since a controlled randomized trial has not been carried out to confirm this. The bootstrap runs underlined the adequacy of the used accuracy intervals and thus provided confidence in the validity of the main analysis. The results are similar to those reported previously [8], in which statistically different PFS values for erlotinib (13.2 months) and gefitinib (9.8 months) were shown, as compared with chemotherapy (5.9 months).
As for the previous pooled analysis [8], there are limitations to this analysis. Statistical comparisons were made in this pooled retrospective analysis between erlotinib, gefitinib and chemotherapy based on PFS. Only high level information like median PFS was obtained from the publications, which could be used to calculate the pooled median PFS. In order to determine accuracy intervals, the simplifying assumption that PFS followed an exponential distribution was necessary but this was not verifiable. However, a bootstrap run confirmed the approximate validity of the accuracy intervals. Also, the schedule of visits for the progression of disease may have differed, according  to the trial protocol. Furthermore, as the composition of the different patient groups (with respect to relevant risk factors) cannot be assessed, the results should be interpreted with due caution. However, there is no indication that the study pool and its treatment subgroups were not representative of the total patient population. Because of the comprehensive nature of the study pool, the differences reported have resulted from data collated from almost the entire body of evidence published up to November 2011. Furthermore, the line of treatment represents an important clinical risk factor that was considered as part of this study pool; the differences between treatments were confirmed through the treatment lines investigated, lending weight to the analyses. These points suggest that, despite the inherent limitations, the differences seen in the study pool deserve attention when the total patient population is considered. Because of the comprehensive approach to pool all available evidence, retrospective studies were included and no quality analysis of the data or mutation testing was possible. Additionally, this pooled analysis compares only median PFS values, and does not include other measures of response or safety/toxicity. Finally, PFS is not always assessed in the same way across all studies, a further source of variability. The present analysis included large, phase III studies that prospectively evaluated the treatments in patients with EGFR mutationpositive NSCLC, which augment the dataset with robust data. The dataset reported here seems to be in agreement with the primary results of the additional phase III trials, which all reported significantly longer PFS with EGFR TKI therapy compared with chemotherapy [3,4,10,111]. The gefitinib data reported here are also in agreement with results of a recently reported phase IV study of gefitinib in Caucasian patients with EGFR mutation-positive NSCLC (n = 106), which observed a median PFS of 9.7 months [112].
There is a need for randomized trials among the EGFR TKIs to directly compare efficacy and toxicity. Trials are ongoing or recently completed, which should provide further data. For example, a randomized, open-label trial recently reported a longer PFS, but slightly more adverse events, with dacomitinib compared with erlotinib in patients with previously treated advanced NSCLC (n = 188) [113]. The LUX-Lung 7 study is a comparative study of afatinib versus gefitinib (NCT01466660) for EGFR mutation-positive NSCLC and is currently recruiting patients. Finally, CTONG 0901 (NCT01024413) was a randomized, phase II trial comparing first-line erlotinib with first-line gefitinib in patients with advanced NSCLC with exon 21 mutations; results have not yet been reported.  Afatinib and icotinib were included in the literature search and pooled analysis. However, there are currently very limited data on these two EGFR TKIs, and statistical analysis of a pooled median PFS could not be accomplished. Further clinical trials are required to establish the role of these agents in the treatment of patients with EGFR mutation-positive NSCLC. Recently, a phase III clinical trial of afatinib was completed and showed that patients receiving afatinib (n = 230) had a median PFS of 11.1 months (compared with 6.9 months with chemotherapy; hazard ratio [HR] = 0.58, 95% CI 0.43-0.78, P = 0.0004) [11]. This study was reported after the pooled analysis was complete, and is therefore not included in the analysis. In a subset of patients with exon 19 and L858R mutations, the median PFS was 13.6 months for afatinib, compared with 6.9 months for chemotherapy (HR = 0.47, 95% CI 0.34-0.65, P < 0.0001). Tolerability of anti-EGFR agents is also important; afatinib had relatively high levels of treatment-related adverse events (diarrhoea: 95%, leading to discontinuation in 1% of patients; rash: 62% and paronychia: 57%).
This pooled analysis utilizes data from a variety of ethnicities, ages and smoking histories, and includes both male and female patients. There are also a variety of EGFR mutations included, although the majority are exon 19 deletions and L858R mutations. The clear efficacy benefits across this range of clinical characteristics confirm the necessity of EGFR mutation testing, rather than reliance on clinical characteristics. Studies in which patients were 'unselected' or selected by clinical characteristics, demonstrate this. The First-SIGNAL study comparing the efficacy of single-agent gefitinib with gemcitabine plus cisplatin as first-line therapy for Korean 'neversmokers' with stage IIIB or IV lung adenocarcinoma not selected by EGFR mutation status [114] was unable to demonstrate superiority for gefitinib over chemotherapy. The IPASS study (phase III, n = 609, gefitinib or carboplatin plus paclitaxel in first line) showed non-inferiority of gefitinib to chemotherapy in non-smokers (or former light smokers) with adenocarcinoma. Only 60% of this preselected population had EGFR mutation-positive tumours. However, a significant PFS benefit of gefitinib compared with chemotherapy was reported in patients with established EGFR mutated-disease (9.5 months versus 6.3 months, respectively, HR = 0.48, 95% CI 0.36-0.64 P < 0.0001) [5]. Obviously, clinical characteristics are not an appropriate surrogate for EGFR mutation testing. Furthermore, although common mutations, such as exon 19 deletions and L858R mutations in exon 21 have been associated with response to EGFR TKIs, many other mutations are detected only occasionally, and correlations with response are not defined. A recent study screened 681 cases and found 18 rare mutations; responses to EGFR TKIs were reported on a case by case basis and varied by mutation [115]. For example, exon 20 and 21 mutations were more likely to confer resistance to erlotinib or gefitinib, while exon 18 and 19 mutations were more often associated with improved efficacy outcome. An analysis of 'other' mutations from the SATURN, TRUST and TITAN trials suggested that some mutations (e.g. in exon 18) conferred a clinical benefit from erlotinib and others (e.g. in exon 20) had a prognostic influence on OS [116]. However, further data are needed.
Erlotinib and gefitinib also have differing responses in the common mutations, and new EGFR TKIs would be expected to also have differences. Exon 19 deletions and L858R mutations have shown similar in vitro sensitivity to gefitinib [22]; however, erlotinib and gefitinib have shown different clinical efficacy depending on whether exon 19 deletions and L858R mutations are present [117,118]. Despite these differences, both drugs have efficacy in patients with both of these mutations and these differences would not influence treatment selection.
As the number of clinical trials evaluating EGFR TKIs continues to increase, the number of patients eligible for pooled analyses such as this one will increase. Updating the dataset will enable more information to be gathered on the effect of TKIs on patients with NSCLC with rare mutations, as well as efficacy outcome of the newer agents. Assessing the benefit of different treatment regimens will also be important, for example, sequential intercalated chemotherapy and erlotinib is being investigated as a promising approach [119].

Conclusions
A comprehensive review of PFS with EGFR TKI therapy or chemotherapy in the treatment of EGFR mutation-positive NSCLC was carried out, and included more than 3500 patients in a pooled analysis. The results demonstrate a clear PFS benefit of treating patients with EGFR mutation-positive NSCLC with an EGFR TKI compared with chemotherapy, with median pooled PFS values of 12.4 months (erlotinib), 9.4 months (gefitinib) and 5.6 months (chemotherapy) reported. This confirms that all patients should be tested for EGFR mutation status immediately on diagnosis of NSCLC, or as soon as feasible. This pooled dataset is in agreement with several large, prospective phase III studies that examined EGFR TKIs as first-line therapy, and strengthens the recommendation that EGFR mutation-positive NSCLC should be treated with erlotinib or gefitinib in the first line. If first-line therapy with EGFR TKIs is not achievable, then consideration should be given to treating with an anti-EGFR agent in any line of therapy, as PFS benefits over chemotherapy are obvious. Further trials should provide more insight into the role of second-generation EGFR TKIs for EGFR mutation-positive NSCLC.