First-line treatments in EGFR-mutated advanced non-small cell lung cancer: A network meta-analysis

Background It remains unknown which is the optimal first-line treatment regimen for patients with advanced epidermal growth factor receptor (EGFR)-mutated non-small cell lung cancer (NSCLC). We performed a network meta-analysis to address this important issue. Methods PubMed, Embase, Cochrane Library, Web of Science and major international scientific meetings were searched for relevant randomized controlled trials (RCTs). Progression-free survival (PFS) data was the primary outcome of interest, and overall survival (OS) and serious adverse events (SAEs) were the secondary outcomes of interests, reported as hazard ratio (HR) or odds ratio (OR) and 95% confidence intervals (CIs). Results 25 RCTs with a total of 5005 patients randomized to receive seven treatments were included in the meta-analysis. Third-generation tyrosine kinase inhibitor (TKI) (osimertinib) and first-generation TKIs (F-TKIs) in combination with chemotherapy (F-TKIs+CT) were more effective than F-TKIs alone in terms of PFS (HR = 0.46, 95% CI: 0.22–0.93; P = 0.031 and HR = 0.62, 95% CI: 0.39–0.98; P = 0.041) and OS (HR = 0.63, 95% CI: 0.43–0.91; P = 0.014 and HR = 0.73, 95% CI: 0.57–0.92; P = 0.008). Second-generation TKIs (S-TKIs) showed significant OS advantage over F-TKIs (HR = 0.83, 95% CI: 0.70–0.99; P = 0.04). Based on treatment ranking in terms of PFS and OS, osimertinib had the highest probability of being the most effective treatment (89% and 86%) and with the best tolerability. F-TKIs+CT was ranked the second-most effective regimen, but with relatively high risk of SAEs. Conclusions Osimertinib seemed to be the most preferable first-line treatment in advanced EGFR-mutated NSCLC. However, limitations of the study including a single RCT investigating osimertinib and immature OS data need to be considered.


Data extraction
Two investigators extracted the following data from each study independently: first author or RCT name, year of publication, duration of RCT, region, interventions, numbers of patients, hazard ratios (HRs) and their 95% confidence intervals (CIs) of PFS and OS, and odds ratios (ORs) and their 95% CIs of serious adverse events (SAEs). Crude HRs with 95% CIs for PFS and OS were either extracted directly from the original reports or calculated by the Kaplan-Meier curves based on the methods of Parmar et al. [27] and Tierney et al. [28]. Data on the overall numbers of patients with SAEs were directly extracted if they were reported in the published article. If only the numbers of individual SAEs were reported separately in articles, we pooled all numbers of them to represent the overall numbers of SAEs.

Quality assessment
Cochrane risk of bias tool [29] was used to evaluate the methodological quality of RCTs, which includes the following five domains: sequence generation, allocation concealment, blinding, incomplete data, and selective reporting. A RCT was rated as "low risk of bias" if all key domains indicated as low risk, was rated as "high risk of bias" if one or more key domains indicated as high risk, and was judged to be "unclear risk of bias" when more than three domains indicated as unclear risk.

Statistical analysis
The primary outcome was PFS, and the secondary outcomes were OS and SAEs. For direct comparisons, standard pairwise meta-analysis (PWMA) was performed. The heterogeneity between studies was assessed by chi-square (χ 2 ) and I-square (I 2 ) tests. A P value <0.10 or I 2 > 50% was considered significant heterogeneity existing, and a random-effects analysis model was used; otherwise, a fixed-effects model was used. PWMA was performed using the software Review Manager 5.3 (Cochrane Collaboration, Oxford, UK). The Bayesian network-meta analysis (NMA) for all outcomes were performed in a randomeffect model [Generalized Linear Model (GLM)] using Markov chain Monte Carlo methods [30][31] in JAGS and the GeMTC package in R (https://drugis.org/software/r-packages/ gemtc). OS and PFS were analyzed with GLM with a normal likelihood incorporating log hazard ratio statistics from individual trials to calculate HR between competing treatments. Count statistics of SAEs was analyzed with GLM with a binomial likelihood to calculate relative treatment effects expressed as OR between different treatments. For each outcome measure, four independent Markov chains were simultaneously run for 20,000 burn-ins and 100,000 inference iterations per chain to obtain the posterior distribution. The traces plot and Brooks-Gelman-Rubin method were used to assess the convergence of model [32]. Treatment effects were estimated by HR/OR and corresponding 95% CI. Network consistency was assessed with node-split models by statistically testing between direct and indirect estimates within treatment loop [33]. To rank probabilities of all available treatments, the surfaces under the cumulative ranking curve (SUCRAs) were calculated [34]. SUCRA equals one if the treatment is certain to be the best and zero if it's certain to be the worst [34]. To jointly compare the efficacy and tolerability of each treatment and to assess their benefit-risk ratios, we ranked them based simultaneously on the SUCRA value of PFS and tolerability (1-SUCRA SAEs ) in the ranking plot. Lastly, comparison-adjusted funnel plot was used to detect the presence of small-study effects or publication bias.
First-line treatments in EGFR-mutated advanced non-small cell lung cancer The remaining 254 studies were screened through a full-text review for further eligibility. All relevant references were also reviewed. Finally, 25 RCTs [4-22, 25, 35-47] with 5005 patients were included in the network meta-analysis and compared seven treatments including F-TKIs, S-TKIs, osimertinib, CT, F-TKIs+CT, F-TKIs+Bev, and F-TKIs+Linsitinib (F-TKIs +Lin). The clinical characteristics are shown in Table 1.

Assessment of included trial
The demographic characteristics of involved patients were generally well-balanced between different trials and different arms within each trial (see S3 Table). Median age ranged from 56 to 68 years. 22-44% of patients were male; and 0-14% were with squamous cell carcinoma. Most of patients were with Stage IV disease (73.2-100%), except patients included in WJTOG3405 trial (47.7%) [6][7]. EGFR mutations were mainly exon 19 deletions and 21 deletions mutations. In eleven studies, details of baseline demographic characteristics were not stated [16-17, 35-38, 40, 44-46]. The risk of bias in included RCTs was summarized in S1 Fig. Four RCTs [4-5, 17, 19, 22] were judged to be unclear risk of bias, as they had more than three domains indicating as unclear risk. The remaining RCTs were judged to be low risk of bias. No trial was rated with a high risk of bias. Funnel plot analysis in term of PFS did not indicate any evident risk of publication bias (S2 Fig). First-line treatments in EGFR-mutated advanced non-small cell lung cancer

Network meta-analysis
The network plot established for NMA is shown in Fig 2. Results of the NMA were presented in Table 3. Osimertinib and F-TKIs+CT were more effective than F-TKIs in terms of PFS

Inconsistency assessment and treatment ranking
There were two independent closed loops in the network for PFS or OS: F-TKIs/S-TKIs/CT and F-TKIs/F-TKIs+CT/CT; one independent closed loop for SAEs: F-TKIs/S-TKIs/CT. Analysis of inconsistency showed that the NMA results were similar to the PWMA results for the three outcomes, which suggested the consistency between the direct and indirect evidence (S3 Fig).
To further assess the benefit-risk ratios of the seven treatments simultaneously, we ranked them based on the SUCRA values of PFS and tolerability (31-SUCRA SAE ) in the ranking plot First-line treatments in EGFR-mutated advanced non-small cell lung cancer (Fig 4). Osimertinib was likely to be the optimal treatment because it had the most efficacy and best tolerability. F-TKIs+CT and F-TKIs+Bev were also two more effective regimens, but with relatively high risk of causing SAEs. S-TKIs achieved relatively good efficacy with moderate tolerability.

Discussion
This novel network meta-analysis assessed the comparative efficacy and tolerability of all major TKIs based first-line treatments (including first-, second-, third-generation TKIs, and combination regimens involving TKIs) in advanced EGFR-mutated NSCLC. It showed that osimertinib provided significant PFS and OS advantage over F-TKIs, and had the highest probability of being the most effective treatment in improving PFS and OS, and with the best tolerability. In the FLAURA study [15] comparing first-line osimertinib with gefitinib or erlotinib in patients with advanced EGFR-mutated NSCLC, the median PFS (mPFS) was significantly longer with osimertinib than with standard F-TKIs (18.9 versus 10.0 months, P < 0.001). Overall SAEs were less frequent with osimertinib than with F-TKIs (34% vs. 45%).
F-TKIs+CT and F-TKIs+Bev were another two more effective regimens in our NMA. Based on treatment ranking, they were ranked second-and third-most effective regimen respectively, but with relatively high risk of causing SAEs. The NEJ009 study, a phase III trial evaluated the efficacy of a combination of gefitinib and CT in advanced NSCLC patients with First-line treatments in EGFR-mutated advanced non-small cell lung cancer EGFR mutations [18]. The combination arm demonstrated significantly improved mPFS (20.9 versus 11.2 months, P < 0.001) and median OS (mOS) (52.2 versus 38.8 months, P = 0.013) compared with gefitinib alone arm. A phase II study [20][21] comparing combination of erlotinib and bevacizumab with erlotinib alone in this patient population reported results with significant benefit on mPFS (16.4 versus 9.8 months, P = 0.0005). Similar improved mPFS was also observed in a phase III trial (NEJ026) [22] (16.9 months in erlotinib+bevacizumab arm versus 13.3 months in erlotinib alone arm, P < 0.001). Although both the two combination regimens were associated with higher incidence of grade 3 toxicities, few patients required dose reduction or withdrawal. The survival difference between F-TKIs and S-TKIs has been investigated in two trials. LUX-LUNG 7 trial [11][12] showed no clinically meaningful survival benefit with afatinib versus with gefitinib in patients with advanced EGFR-mutated NSCLC. However, in another First-line treatments in EGFR-mutated advanced non-small cell lung cancer phase III trial [13][14] comparing dacomitinib with gefitinib in patients with EGFR-mutated NSCLC without brain metastases, dacomitinib was associated with significant improvement in mPFS (14.7 versus 9.2 months, P < 0.001) and mOS (34.1 months versus 26.8 months, P = 0.044) compared with gefitinib, but with increased grade 3 toxicities. In our NMA, S-TKIs was ranked fourth-most effective regimen with moderate risk of causing SAEs.
Based on the findings of this NMA, osimertinib seemed to be the preferable first-line treatment for patients with advanced EGFR-mutated NSCLC. However, the OS data were immature in the FLAURA study [15]. The survival rate at 18 months was not significantly longer with osimertinib than with F-TKIs (83% versus 71%) in the interim analysis [15]. More recently, postprogression outcomes of the FLAURA study have been reported [48]. Median second PFS was not reached [95% CI, 23.7-not calculable (NC)] in the osimertinib arm and 20.0 months (95% CI, 18.2-NC) in the standard-of-care (SoC) EGFR-TKI (gefitinib or erlotinib) arm [HR = 0.58, 95% CI: 0.44-0.78; P = 0.0004]. This suggested that osimertinib preserved clinical benefit after first progression. Moreover, median time to discontinuation of any EGFR-TKI or death was 23.0 months (95% CI, 19.5-NC) in the osimertinib arm and 16.0 months (95% CI, 14.8-18.6) in the SoC EGFR-TKI arm. These exploratory postprogression outcomes showed consistent improvements of osimertinib compared to SoC EGFR-TKI, and provide further confidence in the interim OS data.
There are several limitations in this network meta-analysis. First, in common with other meta-analyses, data were collected and analyzed basis of results reported from trials, and not on individual patient data. Therefore, effects of potential prognostic factors could not be accounted for. Second, most of included trials reported an immature OS data, and follow-up times across trials were different and generally short. Moreover, only one RCT investigated efficacy of osimertinib vs F-TKIs. These limitations do not allow us to reach a definitive conclusion about the superiority of one treatment over another. Third, this network meta-analysis included most of RCTs with Asians (15/25). We combined treatment effects for Asians and other groups assuming that there is no racial difference in the treatment effects. However, there was still no evidence supporting that the effect of TKIs among Asians is comparable to First-line treatments in EGFR-mutated advanced non-small cell lung cancer that among other racial groups, and we could not investigate such a racial difference in the treatment effect using this network meta-analysis data. Fourth, TKIs efficacy may be associated with patient characteristics (such as gender, race, and smoking status), tumor pathology, EGFR mutation types, and developing brain metastasis or not. However, we could not perform the subgroup-analyses because of insufficient data in the individual trials. Finally, some HRs of PFS or OS were calculated from the Kaplan-Meier curve due to that they were not directly reported in the articles. This may result in bias.

Conclusions
Osimertinib seemed to be the most preferable first-line treatment in advanced EGFR-mutated NSCLC. However, limitations of the study including a single RCT investigating osimertinib and immature OS data need to be considered.

Author Contributions
Conceptualization: Jun Dang.