Meta-analysis of efficacy and adverse events of erlotinib-based targeted therapies for advanced/metastatic non-small cell lung cancer

A network meta-analysis evaluating efficacy and adverse events of eight erlotinib-based therapies (erlotinib+placebo, erlotinib+tivantinib, erlotinib+celecoxib, erlotinib+onartuzumab, erlotinib+sunitinib, erlotinib+entinostat, erlotinib+sorafenib, and erlotinib+bevacizumab) for advanced/metastatic non-small-cell lung cancer (NSCLC) was performed. PubMed and Cochrane Library were reviewed, and ten randomized controlled trials were identified in which patients receiving at least one erlotinib-based therapy. Efficacy outcomes, including progression-free survival (PFS), overall survival (OS), overall response rate (ORR), disease control rate (DCR), and adverse outcomes were evaluated. Patients treated with erlotinib+tivantinib, or erlotinib+celecoxib had longer PFS than patients on erlotinib+placebo; patients on erlotinib+tivantinib had longer OS compared to erlotinib+placebo. For PFS, erlotinib+celecoxib had the highest value of surface under the cumulative ranking curve (SUCRA). For OS, erlotinib+tivantinib had the highest SUCRA. For ORR, erlotinib+bevacizumab had the highest SUCRA, while erlotinib+entinostat ranked the lowest. For DCR, erlotinib+sorafenib had the highest SUCRA. Erlotinib+onartuzumab had the highest SUCRA for diarrhea, nausea, vomiting, decreased appetite, and dyspnea. Erlotinib+sunitinib had the lowest SUCRA for diarrhea, nausea, vomiting, and decreased appetite. Erlotinib + entinostat had the lowest SUCRA for fatigue, asthenia, and dyspnea. Our study suggests erlotinib+tivantinib and erlotinib+celecoxib regimens have the best long-term efficacy, while erlotinib+sunitinib and erlotinib+entinostat have the fewest adverse effects in patients with advanced/metastatic NSCLC.


INTRODUCTION
Non-small cell lung cancer (NSCLC) is the most common type of lung cancer, and often develops into an advanced or metastatic disease [1]. Internal genetic and environmental factors, such as smoking, radon, and asbestos have been reported as possible etiologic factors of NSCLC [2,3]. Currently, platinum-based chemotherapy and the application of epidermal growth factor receptor (EGFR) inhibitor erlotinib have been the main strategies for the treatment of advanced/metastatic NSCLC [4]. In NSCLC, c-MET (MET) receptor tyrosine kinase has been associated with the development of resistance to EGFR tyrosine kinase inhibitors (TKIs) in various EGFR-mutant cancers [5,6]. Combined inhibition of VEGFR and EGFR therapy has been used in patients with advanced pretreated NSCLC, using sunitinib plus erlotinib [7].
Erlotinib, a small molecule inhibitor of EGFR, has been used for the treatment of advanced/metastatic NSCLC patients who do not respond to chemotherapy regimens [1,8,9]. Moreover, erlotinib can improve survival in patients with untreated NSCLC who have EGFR-activating

Meta-Analysis
Oncotarget 86817 www.impactjournals.com/oncotarget mutations [10]. Combination treatments for advanced/ metastatic NSCLC are becoming more popular, since they may be more effective than a single drug treatment. Combining erlotinib with sorafenib results in a dual inhibition of EGFR signaling and angiogenesis, which are two vital targets in treatment of NSCLC [11]. However, a combined therapy using entinostat and erlotinib is still controversial, since patients with advanced NSCLC do not respond to the treatment [12]. Several studies have evaluated the efficacy of erlotinib-based targeted therapies, but with mixed results [4,7,13,14]. In addition, agents like erlotinib and bevacizumab may cause inevitable adverse events including rash, diarrhea, dry skin, and fatigue; therefore, it is necessary to determine which targeted therapy is safer and produces less toxicity [15].
A meta-analysis integrates the results of various independent studies, thus increasing the statistical power [16]. In this study, we employed a network meta-analysis approach to compare the efficacy and adverse events among eight targeted, erlotinib-based therapies (regimens of erlotinib + placebo, erlotinib + tivantinib, erlotinib + celecoxib, erlotinib + onartuzumab, erlotinib + sunitinib, erlotinib + entinostat, erlotinib + sorafenib and erlotinib + bevacizumab) for patients with advanced/metastatic NSCLC.

Baseline characteristics of eligible studies
A total of 1,991 published studies were initially identified through electronic databases and manual searches. After excluding duplicates (n = 9), letters, reviews or abstracts (n = 260), non-human studies (n = 199) and non-English studies (n = 245), the remaining 1278 studies were examined. Subsequently, nonrandomized control trials (n = 367), studies not relevant to advanced/metastatic NSCLC (n = 306), studies not relevant to targeted therapies (n = 594) and a study with uncomplete data (n = 1) were also excluded. Finally, 10 randomized controlled trials that met inclusion criteria were enrolled in this meta-analysis [1, 4-7, 10, 11, 15, 17] (Supplementary Figure 1). There were 3,792 patients with advanced/metastatic NSCLC; patients who received erlotinib + placebo, and erlotinib + tivantinib regimens accounted for the majority. All eligible studies were published between 2011 and 2015. One study included Asian subjects while nine studies included European and American subjects. All 10 eligible studies were twoarm trials. Baseline characteristics of eligible studies and the Cochrane risk of bias assessment are presented in Supplementary Table 1

Cumulative ranking probabilities of eight targeted therapies
Cumulative ranking probabilities of eight targeted therapies for advanced/metastatic NSCLC are illustrated in Table 5. The results of SUCRA values demonstrated in the aspect of efficacy, in terms of PFS, the cumulative ranking probability of erlotinib + celecoxib regimen was the highest (83.0%); in terms of OS, erlotinib + Oncotarget 86821 www.impactjournals.com/oncotarget tivantinib regimen had the highest cumulative ranking probability (93.3%); in terms of ORR, the cumulative ranking probability of erlotinib + bevacizumab regimen was the highest (86.4%) while that of erlotinib + entinostat regimen was the lowest (23.9%); in terms of DCR, erlotinib + sorafenib regimen had the highest cumulative ranking probability (75.2%). Regarding adverse events, the cumulative ranking probability of erlotinib + onartuzumab regimen was the highest in diarrhea (88.0%), nausea or vomiting (93.0%), decreased appetite (79.8%) and dyspnea (84.5%); the cumulative ranking probability of erlotinib + placebo regimen was the highest in fatigue or asthenia (76.5%); but the cumulative ranking probability of erlotinib + sunitinib regimen was the lowest in diarrhea (21.8%), nausea or vomiting (27.8%) and decreased appetite (29.5%). The cumulative ranking probability of erlotinib + entinostat regimen was the lowest in fatigue or asthenia (30.8%) and dyspnea (32.3%).
Moreover, the information about ethnicity and prior therapy was included according to PFS and OS indicators, and meta regression analysis was performed for revision of PFS and OS results. SUCRA curves were drawn to resequenced interventions. In terms of PFS-Ethnicity, the result of intervention after correction showed a minor deviation compared with that before correction. Erlotinib + celecoxib regimen before correction showed the highest value of cumulative sort probability, while showed the second after correction (71.52%), which was lower than erlotinib + tivantinib regimen (72.3%). Therefore, it indicated the effect of ethnicity on the patient's survival. However, for PFS-Type of prior therapy, OS-Ethnicity and OS-Type of prior therapy, the result of interventions after correction were in line with that before correction, indicating no obvious heterogeneity. In conclusion, the results of our study are reliable as the heterogeneity was controlled by meta-analysis, which showed important Notes: WMD = Weighted mean difference; OR = Odds ratio; 95% CI = 95% confidence interval, Weighted mean difference or odds ratio (95% CI) below the treatments should be read from row to column while above the treatments should be read from column to row. PFS and OS are stated as WMD (95% CI), while ORR and DCR are presented as OR (95% CI). PFS = Progression-free survival; OS = overall survival; ORR = overall response rate; DCR = Disease control rate; A = Erlotinib + Placebo; B = Erlotinib + Tivantinib; C = Erlotinib + Celecoxib; E = Erlotinib + Sunitinib; F = Erlotinib + Entinostat; G = Erlotinib + Sorafenib; H = Erlotinib + Bevacizumab. www.impactjournals.com/oncotarget significance for targeted therapies in treatment with advanced/ metastatic NSCLC patients (Figure 4).

DISCUSSION
The pairwise meta-analysis suggested that erlotinib + tivantinib, and erlotinib + celecoxib regimens had better long-term efficacy for the treatment of advanced/ metastatic NSCLC. However, the adverse events of diarrhea, nausea or vomiting, decreased appetite, and dyspnea, exhibited the highest SUCRA value in erlotinib + onartuzumab regimen. For diarrhea, nausea or vomiting, and decreased appetite, the erlotinib + sunitinib regimen exhibited the lowest SUCRA; for fatigue or asthenia, and dyspnea, the erlotinib + entinostat regimen exhibited the lowest SUCRA value. The tyrosine kinase c-MET acts as a tumor suppressor in NSCLC, and as a resistance mediator to erlotinib in EGFR-activating mutations [10]. When erlotinib is given with tivantinib, which is a novel, selective inhibitor of c-MET, to patients with advanced/metastatic NSCLC, tivantinib can contribute to prolonged PFS and improved OS [5]. Moreover, erlotinib and afatinib could be the best choice for patients with chemo-naïve EGFR mutations. In addition, erlotinib has a potential survival benefit in patients who have previously been treated [18]. Scagliotti et al. reported that erlotinib plus sunitinib exhibited antitumor activities by inhibiting tumor growth, metastasis, and angiogenesis, and were associated with a longer PFS and greater ORR [7]. However, a previous study showed that treatment-related adverse events from the combination of erlotinib and sunitinib were more frequent than from erlotinib alone, including diarrhea, anorexia, fatigue, nausea, and dyspnea; this is consistent with our results [4]. Combination of bevacizumab and erlotinib has been effective in prolonging the PFS and ORR in patients, while having minimal side effects [1]. Targeting multiple molecular pathways can increase the efficacy and avoid resistance development, without increasing adverse events (AEs) [19]. A phase II study of erlotinib, placebo-controlled and randomized, without and with entinostat, was conducted for treatment of patients with advanced NSCLC [12].
Based on our results and SUCRA values, erlotinib + tivantinib, and erlotinib + onartuzumab regimens had fewer adverse events, while erlotinib + sunitinib had a higher incidence of adverse events for patients with advanced/metastatic NSCLC. Onartuzumab can bind the c-MET extracellular domain to inhibit hepatocyte growth factor binding and activation, thus contributing to the improved PFS and OS in patients with advanced/ Oncotarget 86823 www.impactjournals.com/oncotarget metastatic NSCLC [10]. As for the safety, erlotinib + sunitinib, and erlotinib + entinostat regimens ranked lower, indicating that these two targeted therapies might have more side effects. Previous studies have indicated that entinostat may inhibit epigenetic modifications to reverse the resistance to epidermal growth factor receptor tyrosine kinase inhibitor therapy in advanced/metastatic NSCLC [20,21]. However, erlotinib + onartuzumab with the highest cumulative ranking probabilities indicated a relatively lower incidence of adverse events, which is consistent with our results of network meta-analysis.
Our present study systematically compared the efficacy and adverse events of eight erlotinib-based targeted therapies for advanced/metastatic NSCLC with direct and indirect evidence [22]. However, this network meta-analysis has some limitations. First, the number of randomized controlled trials included in this study was relatively small, which could have an influence on the universality of our results. Second, information of some outcome indicators was not complete; therefore, we did not use the node-splitting method to analyze the inconsistency of outcome indicators and also did not carry out a cluster analysis of the SUCRA values.
There was no statistical difference in short-term efficacy, while erlotinib + tivantinib, and erlotinib + celecoxib regimens had a better long-term efficacy in patients with advanced/metastatic NSCLC. Regarding the adverse events, they were higher for erlotinib + sunitinib, and erlotinib + entinostat regimens; this may provide clinical guidelines for targeted therapies for patients with advanced/metastatic NSCLC. However, the validity of our results may be affected by the fact that seven out of ten included studies did not analyze the c-Met levels. In addition, our results suggest that ethnicity has a significant

Retrieval strategy
The electronic retrieval of English databases, PubMed and Cochrane Library, was conducted from their inception to February 2017, supplemented with manual retrieval of relevant references. The electronic retrieval combined keywords and free words to search for references. The keywords included advanced or metastatic NSCLC, targeted therapies, and randomized controlled trials.

Data extraction and quality assessment
The data contained in the included studies were collected by two researchers independently with the application of unified data collection form. Disagreements between two reviewers were resolved by discussion with other researchers until consensus was achieved. Two or more researchers assessed included studies with the Cochrane risk of bias assessment tool [23]. The assessment included assigning a judgment of "yes," "no," or "unclear" for each domain to designate a low, high, or unclear risk of bias, respectively. A study with no more than 1 "unclear" or "no" domain would be identified as having a low risk of bias; a study with 2-3 "unclear" or "no" domains would be regarded as having an unclear risk of bias; and a study with over 4 "unclear" or "no" domains would be deemed as having a high risk of bias [24]. Review Manager 5 (RevMan 5.2.3, Cochrane Collaboration, Oxford, UK) was used to assess quality assessment and investigate publication bias.

Statistical analysis
Initially, traditional pairwise meta-analyses were performed to compare the eight targeted therapies. The pooled estimates of weighted mean differences (WMD), odds ratios (OR) and 95% confidence intervals (CIs) were calculated. The heterogeneity test of the studies was performed by Chi-square test and I-square test [25]. The R3.2.1 software was used to draw network diagrams of the targeted therapies. Every node represents one targeted therapy; the node size represents the number of the corresponding targeted therapy; the line thickness between two nodes represents the number of paired studies of two targeted therapies. In addition, we carried out Bayesian network meta-analyses to compare eight targeted therapies to each other as well. Each analysis was based on non-informative priors for effect sizes and precision. Convergence and lack of auto correlation were checked and confirmed after four chains and a 20,000-simulation burn-in phase. Finally, direct probability statements were derived from an additional 50,000-simulation phase [26]. To assist in the interpretation of WMDs or ORs, we calculated the probability of each targeted therapy being the most effective or safest treatment method which was based on a Bayesian approach using probability values summarized as the surface under the cumulative ranking curve (SUCRA). A targeted therapy with a larger SUCRA value represents a better efficacy [27,28]. All calculations were done using R (V. 3

CONFLICTS OF INTEREST
The authors report no conflicts of interest in this work.