Network meta analysis of first-line therapy for advanced EGFR mutation positive non-small-cell lung cancer: updated overall survival

Lung cancer is one of the leading causes of morbidity and mortality worldwide. In 2018, 2.1 million lung cancer cases were reported globally, with the highest age-standardized incidence rates found in Micronesia/Polynesia and Eastern Europe [1]. Approximately 85% of lung cancers are characterized as non-small-cell lung cancer (NSCLC) with greater than 50% of patients diagnosed presenting with advanced disease at diagnosis, which is strongly correlated with poor prognostic outcomes. As a result, NSCLC is a leading cause of cancer-related death worldwide [2–4]. Mutations in EGFR are a primary oncogenic driver in NSCLC, accounting for approximately 10–18% and 40–55% prevalence in Caucasians and Asians, respectively [5]. For patients with advanced NSCLC with EGFR-activating mutations (exon 19 deletion or exon 21 L858R substitution mutations), National Comprehensive Cancer Network and the European Society for Medical Oncology Clinical Guidelines recommend first-line treatment with EGFR- targeted tyrosine kinase inhibitors (TKIs), which are both efficacious and well-tolerated relative to chemotherapy agents [6,7].

Several EGFR TKIs are globally and regionally (i.e., Asia or continental Europe only) approved as monotherapy for first-line treatment of advanced EGFR mutation positive (EGFR+) NSCLC. As new EGFR TKIs come to market (dacomitinib and osimertinib), it is important to continually evaluate how the current globally and regionally approved EGFR TKI therapies compare in terms of efficacy. There are no direct comparative randomized controlled trials (RCTs) for all EGFR TKI comparisons. In an attempt to fill this gap, five network meta-analyses (NMAs) [8–12] have been previously published investigating the first-line treatment of patients with locally-advanced or metastatic EGFR+ NSCLC comparing EGFR TKIs and chemotherapy.

Most recently, the NMA published by Franek et al. [12] demonstrated treatment with dacomitinib and osimertinib trended directionally toward improved progression-free survival (PFS) relative to other EGFR TKIs (dacomitinib vs afatinib: hazard ratio [HR]: 0.80; 95% credible interval [CrI]: 0.57–1.13; dacomitinib vs erlotinib: HR: 0.48; 95% CrI: 0.25–0.91; dacomitinib vs gefitinib: HR: 0.59; 95% CrI: 0.47–0.74; osimertinib vs afatinib: HR: 0.61; 95% CrI: 0.43–0.86; osimertinib vs dacomitinib: HR: 0.76; 95% CrI: 0.55–1.04; osimertinib vs erlotinib: 0.36; 95% CrI: 0.19–0.70; osimertinib vs gefitinib: HR: 0.45; 95% CrI: 0.36–0.56) [12]. Numerical improvement of OS was also observed for treatment with dacomitinib (vs afatinib: HR: 0.89; 95% CrI: 0.61–1.29; vs erlotinib: HR: 0.81; 95% CrI: 0.45–1.47 and vs gefitinib: 0.76; 95% CrI: 0.58–0.99).

However, at the time of the original NMA conducted by Franek et al. [12], the OS data for osimertinib from the FLAURA trial [13] were based on the interim analysis, which corresponded to 25% of expected events, and were not included in the NMA for OS. Since OS is a critically important clinical outcome to assess efficacy of cancer treatments [14], this NMA was conducted to update the Franek et al. [12] NMA. The current NMA also includes the updated OS results for dacomitinib from the ARCHER 1050 trial, analyzed after an extended median follow-up of 47.9 months (data cut-off date of 13 May 2019), [15] along with the mature OS results for osimertinib from the FLAURA trial, based on the planned final analysis with median follow-up of 35.8 months in the osimertinib arm (data cut-off date of 25 June 2019) [16]. In addition to comparing OS of first-line EGFR TKI therapies for the treatment of patients with advanced or metastatic EGFR+ NSCLC in the overall trial populations, analyses were conducted for OS in four subgroups including two mutation subtypes (exon 19 deletion and exon 21 L858R substitution mutations) and two subpopulations (Asian vs non-Asian ethnicity).

Materials & methods

The search strategy, selection criteria, data extraction/quality appraisal and statistical methods have been previously described in detail [12]. Briefly, a systematic literature review (SLR) was performed according to the University of York Centre for Reviews and Dissemination guidance for undertaking systematic reviews in healthcare [17] and the results are reported according to the ‘preferred reporting items for systematic reviews and meta-analyses’ reporting checklists for incorporating NMA of healthcare interventions [18,19].

The original search included a systematic search of 11 electronic databases for trials published from 1 January 2004 to 1 August 2018 [12]. In addition, a systematic Embase search for conference abstracts from the past 5 years was conducted for the following conferences: American Society of Clinical Oncology, European Society of Medical Oncology, International Association for the Study of Lung Cancer, World Conference on Lung Cancer, European Lung Cancer Conference and International Society for Pharmacoeconomics and Outcomes Research. For the updated NMA, a manual search was conducted to include any new and relevant publications corresponding to the ARCHER 1050 and FLAURA trials up to November 2019. Only RCTs of first-line EGFR TKI therapies and their comparators that molecularly selected or stratified for patients with EGFR+ NSCLC were included.

A Bayesian NMA [20] was conducted in R (v3.6.1) using the package ‘GeMTC’ (v0.8-2), which calls upon JAGS (using the rjags package [21]) for Markov Chain Monte Carlo simulations using a normal likelihood and identity link [22]. Cox proportional HRs along with their corresponding CrIs, were used as the summary estimates of relative treatment effects. Log HRs and their corresponding standard errors were used as inputs in the fixed-effect models. A minimum of 50,000 inference iterations was used to ensure model convergence with a burn-in of at least 10,000. All models were evaluated for convergence and model fit. However, since the network did not contain closed loops, inconsistency was not assessed.

Rank probabilities for each treatment based on Bayesian NMA were also produced [20,23]. Rank probabilities reflect the number of times a treatment ranked first (or second, third, etc.) out of the total number of random samples. The actual rank in a sample was determined by the mean point estimate of each comparison along with the width and overlap of the 95% CrIs [24]. ‘Statistical significance’ was reached when the CrI’s did not contain the null value of one.

Similar to the Franek et al. publication [12], since there were many nodes with only one connection and variations in the use and approval of the different EGFR TKIs across regions, both a base case (BC) (Figure 1) and a sensitivity network (Supplementary Figure 1) that included regionally approved therapies (icotinib, relevant for China and erlotinib in combination with bevacizumab, relevant for continental Europe) were constructed. In addition, the BC analyses were conducted for OS in four subgroups including two mutation subtypes (exon 19 deletion and exon 21 L858R substitution mutations) and two subpopulations (Asian vs non-Asian ethnicity); see Supplementary Figure 2 for details.

Results

SLR updates

The results of the original SLR have been reported previously [12]. Briefly, after screening was completed in the original SLR [12], 15 first-line therapy RCTs were included. This update identified an additional three publications (two conference presentations and one associated published manuscript) that were eligible for inclusion as they represented updated OS results for the ARCHER 1050 trial and mature OS results from the FLAURA trial [15,16,25].

NMA results

The BC network (Figure 1) included five trials: ARCHER 1050, ENSURE, FLAURA, LUX-Lung 6 and LUX-Lung 7 [3,15,16,26,27]. Using data from the five trials, this network allowed comparisons among the following EGFR TKIs: afatinib, dacomitinib, erlotinib, gefitinib, osimertinib and cisplatin in combination with gemcitabine. An additional sensitivity network added three trials (CONVINCE, JO25567 and LUX-Lung 3), which included the two regionally approved therapies: icotinib and erlotinib in combination with bevacizumab [4,27,28]. The LUX-Lung 3 trial was included in the sensitivity network to connect icotinib into the network. See Supplementary Table 1 for details of trial sample sizes and reported OS results. Trials comparing EGFR TKIs to chemotherapies that otherwise did not connect to the networks were excluded from the NMA [29–32].

Table 1 & Figure 2 represent the relative effects of EGFR TKIs in the BC network. Based on the NMA, dacomitinib demonstrated a significant improvement of OS compared with gefitinib (HR: 0.75; 95% CrI: 0.59–0.95) and a numerical improvement of OS versus afatinib (HR: 0.87; 95% CrI: 0.61–1.24), erlotinib (HR: 0.79; 95% CrI: 0.44–1.42) and osimertinib (HR: 0.94; 95% CrI: 0.68–1.29). Osimertinib showed a numerical improvement of OS versus afatinib (HR: 0.93; 95% CrI: 0.66–1.31), erlotinib (HR: 0.85; 95% CrI: 0.48–1.51) and gefitinib (HR: 0.80; 95% CrI: 0.64–1.00). Similarly, afatinib showed a numerical improvement of OS versus both erlotinib (HR: 0.91; 95% CrI: 0.57–1.45) and gefitinib (HR: 0.86; 95% CrI: 0.66–1.12). Regarding treatment rank probabilities, dacomitinib had the highest probability of being ranked first in the network (50%), followed by osimertinib (25%), erlotinib (15%) and afatinib (9%), see Supplementary Table 2 for details.

The additional sensitivity NMA included regionally approved therapies icotinib and erlotinib in combination with bevacizumab in the network (Table 2). While all other EGFR TKI comparisons remained relatively similar in this sensitivity analysis, erlotinib in combination with bevacizumab showed a numerical improvement of OS relative to most EGFR TKIs (afatinib HR: 0.89; 95% CrI: 0.48–1.66; erlotinib HR: 0.81; 95% CrI: 0.53–1.23; gefitinib HR: 0.76; 95% CrI: 0.39–1.51; icotinib HR: 0.71; 95% CrI: 0.34–1.52 and osimertinib HR: 0.96; 95% CrI: 0.47–1.95). Nevertheless, erlotinib in combination with bevacizumab did not show a numerical improvement of OS versus dacomitinib (HR: 1.02; 95% CrI: 0.50–2.09). Regarding treatment rank probabilities, erlotinib in combination with bevacizumab had the highest probability of being ranked first in the sensitivity network (40%), followed by dacomitinib (34%), osimertinib (16%) and afatinib (4%), see Supplementary Table 2 for details.

The results for the mutation and ethnicity subgroup analyses in the BC network were consistent relative to the overall population analysis (Figure 3 & Supplementary Figure 1 & Supplementary Table 3a–d). Treatment with dacomitinib showed a numerical improvement of OS with point estimates consistently in favor of dacomitinib versus other EGFR TKIs. The exceptions were the comparison of dacomitinib versus afatinib in the exon 19 deletion mutation subgroup (HR: 1.02; 95% CrI: 0.64–1.64), and the comparison of dacomitinib versus osimertinib in the exon 19 deletion mutation (HR: 1.25; 95% CrI: 0.82–1.91) and non-Asian (HR: 1.40; 95% CrI: 0.78–2.50) subgroups, although these results were not statistically significant.

Discussion

We report an updated NMA of RCTs examining the relative effects of first-line EGFR TKIs (afatinib, dacomitinib, erlotinib, gefitinib, icotinib, osimertinib and erlotinib in combination with bevacizumab) for OS in patients with EGFR+ advanced or metastatic NSCLC. Many of the included trials were stratified by predefined subgroups prior to randomization; a strength in this NMA, which also allowed for the examination of EGFR mutation and ethnicity subgroup effects. Overall, dacomitinib showed a consistent numerical improvement of OS relative to other EGFR TKIs across the BC networks for the overall population and for subgroup analyses. In addition, dacomitinib had the highest probability of being ranked first for OS among the other EGFR TKIs. However, the CrIs of many comparisons were wide and often did not represent a statistically significant result.

Slight differences in the OS results for the exon 19 deletion mutation and exon 21 L858R substitution mutation subgroup analyses were observed. The exon 21 L858R substitution results were consistent with the overall population results, whereas, in the exon 19 deletion mutation subgroup, afatinib and osimertinib demonstrated a numerical improvement of OS relative to dacomitinib. In the Asian subgroup analyses, individuals treated with dacomitinib were shown to have a numerical improvement of OS relative to the other EGFR TKIs similar to the overall population. However, treatment with osimertinib in the non-Asian subgroup demonstrated a numerical improvement of OS relative to the other EGFR TKIs.

It is important to note that these results should be interpreted with caution given the lack of statistical significance found in most of the estimates. This is particularly the case for the non-Asian subgroup. Non-Asian patients were not recruited in the ENSURE and LUX-Lung 6 trials, and thus the network for the non-Asian subgroup included only three trials, with small sample sizes of non-Asian patients within each trial. Finally, a sensitivity analysis network was constructed to test the robustness of the models by including treatments relevant only to clinical practice in certain geographic regions. Consideration of regional practice did not substantially change the results of the BC network.

This updated NMA, which synthesizes the most recent evidence, included the most up to date and mature OS data for dacomitinib and osimertinib from the ARCHER 1050 and FLAURA trials, respectively [15,16]. The stringent inclusion criteria were a strength of this NMA, however, only five trials were included in the BC network. Therefore, uncertainty among these results is high due to a limited number of trials and the subsequently small sample sizes included.

Furthermore, since the CrIs overlapped between subgroups, the findings for these populations are limited (as noted in Franek et al. [12]). The only exception was the comparison of dacomitinib with gefitinib. The CrI for the relative effect estimates for dacomitinib versus gefitinib were narrow as these were derived directly from the ARCHER 1050 trial [15], and thus, benefited from increased precision (all other comparisons required increasing steps of indirect evidence).

In addition, most of the trials included in the NMA were not designed to primarily analyze the OS of the specific subgroups, with the exception of the LUX-Lung 6 trial, which comprised an entirely Asian population. The smaller sample sizes of the subgroup populations may not have been powered to detect a statistically significant difference and therefore, may be prone to Type II error. As such, this may have also contributed to heterogeneity between the included trials. Future research should include trials that are specifically designed to examine these specific subgroups (e.g., mutation and ethnicity subgroups) to account for the limitations described above.

Moreover, the results of the included trials did not focus on treatment received postprogression (i.e., treatment after first-line), which may have affected the OS results, especially as recently published trials have included more effective treatment options for patients with EGFR+ NSCLC. For example, in the ARCHER 1050 trial [15], 49.8 and 62.2% of the patients randomized to dacomitinib or gefitinib, respectively, received subsequent systemic therapy after discontinuation of study treatment. Chemotherapy was the most common first subsequent therapy (received by 27.8% of patients randomized to dacomitinib and 35.6% of patients randomized to gefitinib), followed by third-generation EGFR TKIs (dacomitinib: 9.7%; gefitinib: 11.1%) and other EGFR TKIs (dacomitinib: 8.8%; gefitinib: 8.4%). In the FLAURA trial [16], 47.7 and 65.0% of patients randomized to osimertinib or gefitinib/erlotinib received a subsequent therapy after discontinuation of study treatment. Among patients randomized to osimertinib, the most common first subsequent therapy was cytotoxic chemotherapy (received by 32.3% of patients randomized to osimertinib) followed by EGFR TKI other than osimertinib (13.6%). Among patients randomized to gefitinib/erlotinib, the most common first subsequent therapy was EGFR TKI other than osimertinib (received by 48.4% of patients randomized to gefitinib/erlotinib), followed by osimertinib (30.7%) and cytotoxic chemotherapy (14.1%). More robust analyses in the future should consider accounting for postprogression therapies, and perhaps analyzing specific survival time frames (e.g., 3-year survival rate) to help limit the influence of postprogression therapies when examining the efficacy of first-line therapies and OS in EGFR+ NSCLC populations.

Three recent NMAs examining first-line EGFR TKIs and OS in NSCLC have been published and are relevant for comparison [9,11,12]. Results were largely consistent across the three NMAs and this current analysis, although point estimates and CrIs varied slightly due to differences in inclusion criteria and also in natural sampling variation. Specifically, there were no significant differences in results regarding the comparisons among afatinib, erlotinib and gefitinib, with a consistent trend toward benefit of afatinib versus erlotinib and gefitinib observed.

This study included trials identified from 1 January 2004 to 1 August 2018 and publications related to these trials published up to November 2019. However, there have been more recent RCTs that should also be noted for future investigations examining first-line therapies for EGFR+ NSCLC populations. First, a recent Phase III RCT [33] randomized patients with advanced EGFR+ NSCLC to receive either gefitinib alone or gefitinib in combination with pemetrexed and carboplatin chemotherapy. Both PFS and OS were significantly prolonged for patients receiving gefitinib plus pemetrexed and carboplatin, relative to gefitinib alone (median OS not reached for gefitinib plus pemetrexed and carboplatin vs 17 months for gefitinib alone; HR for death, 0.45 [95% CI: 0.31–0.65]). Second, the ongoing Phase III RELAY trial [34], examined the combination of ramucirumab plus erlotinib versus placebo plus erlotinib in stage IV EGFR+ NSCLC patients. This trial demonstrated a prolonged PFS for patients who received ramucirumab plus erlotinib relative to placebo plus erlotinib. Overall survival data were not reported. Finally, two new trials compared erlotinib plus bevacizumab versus erlotinib alone in patients with advanced or metastatic EGFR+ NSCLC [35,36]. In the NEJ026 trial [35], despite a superior PFS for erlotinib plus bevacizumab relative to erlotinib alone, there was no evidence of an OS benefit. However, in an ongoing Phase II US-based trial [36], erlotinib plus bevacizumab did not result in a significant PFS improvement, relative to erlotinib alone.

Further, results comparing dacomitinib and osimertinib with afatinib, erlotinib and gefitinib were also similar between the NMAs, apart from osimertinib being statistically significantly favored over afatinib and erlotinib in the Holleman et al. review [11]. However, in this current NMA, these comparisons were not statistically significant. This difference in the statistical significance of results is likely due to Holleman et al. [11] including immature OS results from FLAURA (HR of 0.63 based on the immature OS data [13] vs 0.80 based on the mature OS data [15,16]) and incorporating two estimates for the FLAURA trial in the NMA. In turn, this may have influenced the weighting of the FLAURA trial in the model. As a result, stronger effect estimates and a more connected network were reported, relative to the network and results in this NMA. For these reasons, the Holleman et al. [11] results should be interpreted with caution.

The landscape for treatment of EGFR+ NSCLC is evolving as additional data have become available for OS for the newly approved EGFR TKIs, dacomitinib and osimertinib. While these analyses provide evidence to support the efficacy of dacomitinib and osimertinib relative to other comparators, future investigations should account for the effects of safety/toxicity, quality of life, treatment resistance and discontinuation rates, subsequent therapies and other comorbidities on OS to improve patient outcomes. Overall, this NMA provides a methodologically rigorous quantitative update of a previously published NMA and elicits new insight into first-line treatment of advanced or metastatic EGFR+ NSCLC with EGFR TKIs.

Conclusion

Overall, this NMA found that dacomitinib had a numerical improvement of OS when compared with other EGFR TKIs in patients with advanced or metastatic EGFR+ NSCLC. Among subgroups, dacomitinib showed a numerical improvement of OS compared with all other EGFR TKIs among patients with exon 21 L858R substitution mutation and among Asian patients while osimertinib showed a numerical improvement of OS compared with all other EGFR TKIs among patients with exon 19 deletion mutation and among non-Asian patients. However, due to the limitations of the analyses discussed above, future research is required to determine if ethnicity and EGFR activating mutations are true effect modifiers for EGFR TKIs. In conclusion, the updated analyses provide additional evidence that dacomitinib should be considered as a first-line treatment option for patients diagnosed with advanced or metastatic EGFR+ NSCLC.