Indirect analysis of first-line therapy for advanced non-small-cell lung cancer with activating mutations in a Japanese population

Lung cancer accounts for more new cases of cancer worldwide (2.1 million cases in 2019) with higher rates of mortality (1.8 million deaths) than any other type of cancer [1,2]. Non-small-cell lung cancer (NSCLC) is the most prevalent type of lung cancer, accounting for 85% of new lung cancer diagnoses, with the majority of patients (75%) being diagnosed with advanced disease [1]. As a result, most lung cancer mortality is attributable to NSCLC [1,3,4]. For NSCLC, one of the driving factors for tumor proliferation is a mutation to the EGFR gene, which directly impacts how the tumor responds to therapy [5]. Asia–Pacific patients with advanced NSCLC have particularly high rates of EGFR-tyrosine kinase inhibitors (TKI)-sensitizing mutations (30–50%), relative to Caucasian patients (10–15%) [6,7]. Among Japanese patients, these mutations have been estimated to be 30.2% for those with NSCLC [8]. The most common EGFR-activating mutations are deletion in exon 19 or exon 21 L858R substitution mutation, which account for approximately 85% of EGFR mutations [9]. In patients with NSCLC and EGFR-activating mutations, EGFR TKIs have been shown to be tolerated well and produce better progression-free survival (PFS) than traditional platinum-based chemotherapy regimens [10–16]. Despite these positive treatment effects, most patients experience disease progression, often when a EGFR TKI-resistant mutation (T790M) has occurred, leading to uncontrolled proliferation, invasion and metastases [17,18]. T790M mutation, a less common EGFR mutation at diagnosis, is acquired in approximately 50–60% of patients treated with a first- or second-generation EGFR TKI [19–21]. Therefore, new therapies are needed to delay the time to therapy resistance and prolong PFS.

Dacomitinib is a second-generation irreversible EGFR TKI that acts on all three members of the ErbB family (HER1, HER2 and HER4) which are often overexpressed/mutated in many forms of cancer [22], while osimertinib is a third-generation irreversible EGFR TKI that inhibits both EGFR TKI-sensitizing and EGFR T790M resistance mutations [23]. In the ARCHER 1050 trial [22], a recent Phase III randomized controlled trial (RCT), dacomitinib was compared with gefitinib for the first-line treatment of advanced EGFR mutation-positive (EGFR+) NSCLC. Dacomitinib demonstrated a statistically significant improvement in PFS by independent radiological central review both in the overall trial and in the Japanese subset (overall trial: hazard ratio [HR] 0.59; 95% CI: 0.47–0.74; p < 0.0001; Japanese subset: HR: 0.54; 95% CI: 0.308–0.946; p = 0.0141) [22,24]. The median PFS for the overall and Japanese subset trials were 14.7 and 18.2 months in the dacomitinib group and 9.2 and 9.3 months in the gefitinib group [22,24]. Similarly, osimertinib was compared with gefitinib or erlotinib (first-generation EGFR TKIs) in the overall FLAURA trial and gefitinib only in the Japanese subset and showed a significantly longer PFS by investigator assessment (overall trial: HR: 0.46; 95% CI: 0.37–0.57; p < 0.001; Japanese subset: HR: 0.61; 95% CI: 0.38–0.99; p = 0.0456), with a median PFS of 18.9 and 19.1 months for osimertinib and 10.2 and 13.8 months for the gefitinib or erlotinib group, respectively [23,25]. Following the publication of these results, dacomitinib and osimertinib (in addition to afatinib, erlotinib, gefitinib and the combination of erlotinib and bevacizumab) have been included in treatment guidelines as a first-line therapy, including the Japanese Lung Cancer Guideline [17]. Osimertinib is also indicated for the treatment of patients with metastatic T790M-positive NSCLC who have disease progression during or after EGFR TKI therapy based on the results of the AURA program [23,26–28]. Due to complexity of conducting direct comparisons among the various EGFR TKIs, there is no definitive information for the positioning of the available EGFR TKIs for first-line treatment of advanced EGFR+ NSCLC in the Japanese subpopulation according to the Japanese Lung Cancer Guidelines [17]. In addition, the FLAURA trial reported slight differences in median PFS improvements when comparing Asian (in general) and Japanese subgroups [23,25].

To compare the efficacy of the available EGFR TKIs for the treatment of EGFR+ NSCLC in Japanese patients (afatinib, dacomitinib, erlotinib, gefitinib, osimertinib and the combination of erlotinib and bevacizumab), a Bayesian network meta-analysis (NMA) was conducted using within-trial HR estimates for PFS. This NMA builds on the previous work [29], but focuses on PFS outcomes among the Japanese subpopulation. This NMA will provide comparisons between the first-line therapies in a relatively homogeneous clinical population. Similar to the previous work [29], this study only included trials where: patient’s EGFR+ status was molecularly determined prior to randomization and; data from populations with mixed or unknown EGFR mutation status were not included.

Materials & methods

We conducted a systematic literature review (SLR), according to the University of York Centre for Reviews and Dissemination (CRD) guidelines for conducting systematic reviews in healthcare [30]. We then reported the results using reporting checklists for including NMA for healthcare interventions from the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) [30–32].

Search strategy

This review was conducted first by searching 11 electronic databases, including: MEDLINE^® In-Process, MEDLINE^® Daily, MEDLINE^®, Embase, EconLIT, Cochrane Library, Cochrane Database of Systematic Reviews (Cochrane Reviews), Database of Abstracts of Reviews of Effects (DARE), Central Register of Controlled Trials Library (CENTRAL), NHS Economic Evaluation Database (NHS EED), Health Technology Assessment (HTA) Database, Clinicaltrials.gov and publicly available published information from National Institute for Health and Care Excellence (NICE) and Scottish Medicines Consortium (SMC). To supplement the peer reviewed papers, we also performed a systematic Embase search for relevant conference proceedings over the past 5 years, including: American Society of Clinical Oncology (ASCO), European Society for Medical Oncology (ESMO), International Association for the Study of Lung Cancer (IASLC), World Conference on Lung Cancer (WCLC), European Lung Cancer Conference (ELCC), International Society for Pharmacoeconomics and Outcomes Research (ISPOR) and ISPOR Annual European Congress. In all cases, searches were limited to articles available in English, published between 1 January 2004 and 3 June 2019 [33]. These dates were selected to correspond with the emergence of literature around EGFR mutations and the effectiveness of TKIs for patients with EGFR+ NSCLC [34].

Selection criteria

The search strategy was designed to find trials that met specific criteria: reported results for adult patients of Japanese ethnicity with locally advanced (stage IIIb) or metastatic (stage IV) EGFR+ NSCLC who had not been previously treated with systematic cancer therapies; EGFR TKI therapy used as the intervention, including: afatinib, dacomitinib, erlotinib, gefitinib, icotinib, osimertinib or erlotinib in combination with bevacizumab; a comparator arm, incorporating any of the following: another EGFR TKI, chemotherapy (if compared with one of the included EGFR TKI interventions), best supportive care/placebo, radiotherapy in combination with TKI or chemotherapy (not as a monotherapy); PFS was reported as an efficacy outcome; and RCT was the study design. Among these, trials were only included if patients were selected or stratified based on their molecularly identified EGFR+ NSCLC prior to randomization in the trial.

Articles were reviewed by two independent reviewers, first by screening the title and abstract, and second by reviewing the full-text articles. A third independent reviewer resolved any conflicts between the two primary reviewers. For all identified review articles, the reference lists were also searched and any relevant citations were included in this SLR.

Data extraction & quality assessment

For the included trials, patient, treatment and trial characteristics, as well as PFS outcomes were extracted by two independent reviewers and the risk of bias was evaluated for all trials using the Cochrane Risk of Bias 2.0 (RoB 2.0) tool for RCTs [35]. The data extracted from each trial included: trial name, authors, year of publication, treatment arm, number randomized, median/mean age, percent males, Eastern Cooperative Oncology Group performance status (ECOG PS), metastasis sites, disease stage, smoking status, percent with deletion 19 and L858R mutations, median duration of follow-up, and HRs and corresponding CIs for PFS (both Investigator Assessed [IA] and Independent Review Committee [IRC], when available). The primary outcome was PFS by IRC, for trials that did not report both the IRC and IA estimates of PFS, the available estimate was used.

Statistical methods

The trials identified through the SLR were assessed to determine the feasibility of including them in an NMA based on subject matter expertise, data inspection and the production of network diagrams. Trial methods and patient characteristics were well balanced, with no reported effect modification seen in the included trials. With these factors in mind, a traditional Bayesian NMA was performed in accordance with NICE and ISPOR technical recommendations [36,37].

Statistical analyses were performed in R (v3.6.1), using the package ‘GeMTC’ (v0.8-2), which includes WinBUGS (using the ‘rjags’ package) [38] for Markov Chain Monte Carlo (MCMC) simulations. The models were fit using a Normal likelihood and identity link [39]. HRs and their corresponding credible intervals (CrI) were calculated to summarize the relative treatment effects for EGFR TKI comparisons. Models were evaluated for convergence (Brooks–Gelman–Rubin method, visual inspection of trace and density plots) and model fit (deviance information criterion, residual deviance and leverage) [40,41]. The networks contained no closed loops of evidence and so inconsistency could not be assessed. Both fixed and random effects models were conducted, but fixed effects models were selected for the primary model as they produced more plausible variance estimates due to the low number of trials, the small sample sizes within each trial and the low connectively in the network diagrams.

Bayesian NMA provides rank probabilities for each treatment in addition to the relative treatment effects (HRs). Rank probabilities reflect the number of times a treatment ranked first (or second, third, etc.) out of the total number of random samples. Rank probabilities were determined by the mean point estimate of each comparison and the width and overlap of the 95% CrIs [42]. The proportional hazards assumption for each trial was assessed [43] after digitization of the Kaplan–Meier curves from each trial [44]. Since one trial [45] did not meet this assumption, a sensitivity NMA was performed using accelerated failure time (AFT) models to generate 1/acceleration factor estimates (instead of HRs) [46]. The interpretation of 1/acceleration factor estimates are similar to HRs in terms of directionality, allowing this sensitivity analysis to be used to confirm the directionality of the results of the base case (BC) NMA and to assess the impact of potential non-proportionality bias on the analysis.

Results

Search results

The SLR identified 31,885 citations (Figure 1), of which 14,842 duplicate citations were excluded, leading to a total of 17,043 citations that underwent screening by title/abstract. The first stage of screening identified 489 citations eligible for full-text review, of which seven RCTs were included, representing n = 1060 patients with EGFR+ NSCLC [10,11,24,45,47–54].

Network construction

The seven RCTs [10,11,24,25,45,53,54] formed a network with several connections between EGFR TKIs and chemotherapies, however not all of the comparisons formed a fully connected network without collapsing some of the nodes. In particular, the erlotinib versus erlotinib in combination with bevacizumab nodes [45,54] were not connected to the other EGFR TKI nodes in the network. Therefore, based on the assumption that gefitinib and erlotinib have equivalent efficacy [29,55,56], these two EGFR TKIs were collapsed into a single node. This single gefitinib/erlotinib node allowed for the inclusion of the erlotinib versus erlotinib in combination with bevacizumab comparisons in the network [45,54]. In addition, it was assumed that the platinum-based chemotherapy regimens across trials had equivalent efficacy, which incorporated all chemotherapy comparison trials in the network [10,11,53]. The final network construction is described as the BC network as shown in Figure 2.

For our BC network, all seven trials were included: ARCHER 1050, FLAURA, JO25567, LUX-Lung 3, NEJ002, NEJ026 and WJTOG3405 [10,11,24,25,45,53,54] (Supplementary Table 1). The traditional Bayesian and AFT NMAs utilized this network. The following EGFR TKIs were included in the BC network: afatinib, dacomitinib, erlotinib, gefitinib, erlotinib in combination with bevacizumab, and osimertinib. As a sensitivity analysis, we created the network using five trials and analyzed gefitinib and erlotinib separately, producing similar results (data not shown).

Patient & trial characteristics

Key patient characteristics were similar across the seven trials included in the BC NMA, as well as between treatment arms within each trial (Table 1). The median age in each trial ranged from 64 to 68 years old and the percentage of males enrolled ranged from 30 to 49%. The majority of trial participants had a ECOG PS of 0–1 (range: 98–100%) [24], and reported that they had never smoked (range: 48–71%). In addition, the presence of exon 19 deletion and exon 21 L858R mutations in each trial ranged from 43 to 65% and 35 to 57%, respectively. At the time of the SLR, patient characteristics (stage and mutation status) for the ARCHER 1050 trial were extracted from a confidential Pfizer Study Report, which has recently been published [57].

Quality assessment

Overall, there was minimal evidence of deviation from intended interventions, missing data estimates, biases in the measurement of outcome(s) and biases regarding the selection of the reported results (Supplementary Table 2). Most trials were not blinded for treatment allocation or for PFS assessors as to treatment allocation; however, PFS may be less prone to this bias as it is based on objective quantitative measures. Furthermore, appropriate randomization methods, outcomes and statistical analyses were performed across the majority of the trials.

NMA results

In the BC NMA, dacomitinib showed a trend toward improved PFS versus all comparators (Table 2). These improvements in PFS were statistically significant for dacomitinib versus gefitinib/erlotinib (HR: 0.54; 95% CrI: 0.31–0.96) and versus platinum-based chemotherapy (HR: 0.21, 95% CrI: 0.11–0.39), as the 95% CrIs did not contain the null value of 1.0. Though the CrIs were wide, dacomitinib also showed a trend toward improved PFS versus all other comparators: versus afatinib (HR: 0.74; 95% CrI: 0.31–1.77), versus erlotinib in combination of bevacizumab (HR: 0.94; 95% CrI: 0.51–1.75), and versus osimertinib (HR: 0.88; 95% CrI: 0.43–1.86). When examining the probability of each EGFR TKI being ranked first (Table 3), dacomitinib had the highest probability (43.3%) of being ranked first, followed by erlotinib in combination with bevacizumab (22.5%) and osimertinib (21.7%).

The AFT NMA produced similar results to the BC NMA. Specifically, PFS showed statistically significant improvement for dacomitinib versus gefitinib/erlotinib and versus platinum-based chemotherapy. This model also showed a trend toward improved PFS for afatinib, erlotinib in combination with bevacizumab and osimertinib. Please see Supplementary Table 3 for further details from the AFT NMA. This sensitivity analysis also indicated that the violation of the proportional hazards assumption in the JO25567 trial [45] would not have largely impacted the results of the BC NMA.

Discussion

This study presents the results of an SLR and Bayesian NMA of RCTs in Japanese patients with EGFR+ NSCLC, used to compare the relative effects of first-line treatment EGFR TKIs on PFS, including: afatinib, dacomitinib, erlotinib, gefitinib, osimertinib and the combination of erlotinib and bevacizumab. Overall, dacomitinib had a consistent trend toward improved PFS relative to all comparators. In particular, there was a marked increase in PFS when comparing dacomitinib to gefitinib/erlotinib or platinum-based chemotherapy. In addition, dacomitinib consistently had the highest probability of being most efficacious among the treatments compared for PFS. However, many of the CrIs for the relative effect comparisons were wide and often spanning the null value. This is likely a result of the small number of trials (n = 7), the small sample sizes within those trials (n = 1060) and the low network connectivity. The small sample sizes resulted partially because Japanese patients were in some cases only included as a subgroup within the full trial. As a result, sufficient power to find differences among comparators that were not already directly compared in trials may have been lacking.

The results of this NMA build on a previous NMA that focused on the global population of patients with EGFR+ NSCLC [29]. The previous NMA [29] found that dacomitinib had significantly improved PFS compared with erlotinib (HR: 0.48; 95% CrI: 0.25–0.91) and gefitinib (HR: 0.59; 95% CrI: 0.47–0.74) and a trend toward improved PFS when compared with afatinib (HR: 0.80; 95% CrI: 0.57–1.13). As such, the results from this NMA, focusing on patients of Japanese ethnicity, have provided similar results to the previous NMA by Franek et al. [29]. However, in the overall population NMA [29], dacomitinib did not demonstrate a trend toward improved PFS compared with osimertinib (HR: 1.31; 95% CrI: 0.96–1.81), in contrast to our results (HR: 0.88; 95% CrI: 0.43–1.86), here. This particular finding highlights the differences between dacomitinib and osimertinib, which, although not statistically significant, does provide potential evidence of treatment effect modification by ethnicity. Furthermore, our results were generally consistent with the Asian subgroup analysis reported in Franek et al. [29]. Similar to the Asian subgroup in Franek et al. [29], the Japanese population results showed a trend toward improved PFS following treatment with dacomitinib when compared with osimertinib in contrast to the overall population results. This potential difference in results suggests improved efficacy of dacomitinib compared with osimertinib in Asian, and specifically Japanese, patients with EGFR+ NSCLC.

Characteristics including but not limited to ethnicity, sex, smoking status and presence of brain metastases at diagnosis have been examined as independent prognostic factors in patients with EGFR+ NSCLC [58–61]. This review adds important evidence specific to patients with Japanese ethnicity (and more broadly, Asian ethnicity) whom may possess different biological underpinnings of NSCLC relative to non-Asian patients [58]. However, we were not able to conduct any additional subgroup analyses within the Japanese population due to limited subgroup information provided in the publications. In the ARCHER 1050 and FLAURA trials [24,25], it was noted that slight imbalances in patient characteristics were observed in the Japanese population. Specifically, there was an imbalance in the percentage of males observed in the ARCHER 1050 and FLAURA trials, with a higher percentage of males in the gefitinib arm [24,25] relative to the other gefitinib arm trials (NEJ002 and WJTOG3405) [10,11]. Therefore, sex may be a potential effect modifier in the Japanese population and warrants further investigation.

When interpreting these results, there are limitations that should be noted. First, an underlying assumption of our methodological approach is that the included trials and outcomes are sufficiently similar to allow for data to be pooled. Specifically, the FLAURA and the LUX-Lung 3 trials [25,53] stratified their randomization by Asian versus non-Asian ethnicity but not specifically by Japanese ethnicity and as a result, Japanese ethnicity could be a potential source of bias on the treatment effects [62]. Another limitation to note for this NMA, was that in order to analyze a fully connected network, erlotinib and gefitinib were collapsed as one treatment node. While there was evidence to support the assumption of equivalent efficacy among these TKIs [29,55,56], this is a limitation that should be taken into account when interpreting these results. Moreover, the findings from this NMA are limited to indirect comparison of EGFR TKIs for first-line therapy in advanced or metastatic NSCLC patients with the most common EGFR mutations (exon 19 deletion and exon 21 L858R substitution mutation). Five out of the seven RCTs included in this NMA included only patients with these mutations, and two RCTs (LUX-Lung 3 and NEJ002) had less than 10% of patients with uncommon mutations (Table 1). Among patients with uncommon EGFR mutations of exon 18–21, the Japanese Lung Cancer Society Guideline recommends first-line treatment with gefitinib, erlotinib or afatinib for patients with E709X, G719X, S768I, P848L, L861Q mutations; EGFR TKIs are not considered as a first-line option for patients with exon 20 insertion [17]. Additionally, the common mutation estimate from the LUX-Lung 3 trial was used throughout this NMA for consistency across all included trials [53] and the results of the NMA may vary if a different mutation estimate was used. Further, inclusion criteria in this review did not account for the presence of brain/CNS metastases and there were three trials that included patients with brain/CNS metastases [25,53,54] and four that did not [10,11,24,45]. On another note, other RCTs examining first-line therapies in patients with EGFR+ NSCLC were not included in this analysis due to not meeting the inclusion criteria of having results presented for a Japanese subgroup of patients [3,4,13,15,16,63–66]. Finally, platinum-based chemotherapy regimens were collapsed into one node [55], but it is important to recognize that cisplatin regimens have demonstrated higher clinical efficacy than carboplatin regimens. Results should be interpreted with these limitations in mind.

This work also has important strengths. First, to our knowledge, this is the first NMA conducted in Japanese patients with EGFR+ NSCLC examining first-line TKIs for treatment efficacy. The overall quality of evidence in the SLR was high, as RCTs were only included if they used an intention-to-treat analysis, detailed outcomes were reported, and patients were prospectively selected or stratified based on EGFR+ or mutation status prior to randomization. In addition, most trials used PFS determined by IRC. The SLR represents a relatively homogeneous trial population when comparing patient characteristics and the outcomes presented among the included trials. Results from this NMA provide indirect comparisons among all of the available EGFR TKIs among Japanese patients and could be informative to refine recommendations specifically around the sequencing of different first-line EGFR TKIs in further treatment guideline development for Japanese patients with EGFR+ NSCLC.

Conclusion

Dacomitinib demonstrated a trend toward improved PFS and was consistently the most efficacious relative to all comparators, although CrIs were wide. Therefore, dacomitinib could be considered one of the first-line therapy options for Japanese patients diagnosed with EGFR+ NSCLC. More definitive data from RCTs with direct comparisons of different first-line treatment EGFR TKIs are needed to confirm the results of this NMA. Future investigations should also consider examining safety/toxicity, quality of life, treatment resistance, subsequent therapies and other comorbidities to improve patient outcomes for advanced or metastatic EGFR+ NSCLC.

Future perspective

Dacomitinib shows a trend toward improved PFS relative to all EGFR TKI comparators for first-line use in patients of Japanese ethnicity and EGFR+ NSCLC. These indirect comparisons are based on a relatively low number of trials with small sample sizes, and so it is important to monitor the comparative efficacy of first-line EGFR TKIs within the Japanese population. Direct comparisons are needed to confirm the optimal sequencing of the different first-line EGFR TKIs and ensure improvement in patient outcomes.