A comparison of the efficacy of antiangiogenic agents combined with chemotherapy for the treatment of non-small cell lung cancer: a network meta-analysis

To explore the effects of combinations of antiangiogenic agents and chemotherapy agents on non-small cell lung cancer (NSCLC) patients and indirectly compare the therapeutic effect of Endostar combined with chemotherapy and bevacizumab combined with chemotherapy on NSCLC. We searched 3 electronic databases: PubMed, Web of Science and the Cochrane Library. The ORRs, HRs and 95% confidence intervals of OS and PFS were used to compare the efficacy of Endostar combined with chemotherapy and bevacizumab combined with chemotherapy. We use the Bayesian network meta-analysis method to make indirect comparisons and obtain rank probabilities; in addition, we used single-arm meta-analysis to synthesize the existing data. A total of 29 studies were included in the analysis. Among them, we included a total of 14 interventions. A total of 12,862 patients participated in this analysis. The single-arm meta-analysis showed that the pooled ORR and 95% CI were 0.35 (0.31, 0.39), the pooled HR of OS and 95% CI were 0.89 (0.81, 0.98), and the pooled HR of PFS and 95% CI were 0.67 (0.56, 0.81). According to the results of network meta-analysis, there were no significant differences between the 5 kinds of bevacizumab combined with chemotherapy regimens and the 4 kinds of Endostar combined with chemotherapy regimens for improving ORR and prolonging OS and PFS. The rank probabilities suggested that in terms of ORR, Pla + Pem + Bev was the first-ranked intervention (0.288). Pla + Pem + Endo was the first-ranked intervention for prolonging OS (0.423) and Pla + Gem + Endo was the first-ranked intervention for prolonging PFS (0.302). Antiangiogenic agents combined with platinum-containing dual drugs can provide benefits to NSCLC patients. In addition, bevacizumab combined with chemotherapy regimens has better theraputic effect on ORR while Endostar combined with chemotherapy may have better effects on OS and PFS for the treatment of NSCLC patients.

squamous cell carcinoma, adenocarcinoma, large cell carcinoma and adenosquamous carcinoma. Angiogenesis, which was proposed by Folkman in 1971 [2] and listed as one of the hallmarks of tumors by Hanahan [3], plays an important role in the occurrence and development of tumors. Angiogenesis is related to the proliferation, invasion and metastasis of tumors [4]. Many molecules, such as vascular endothelial growth factor (VEGF), platelet-derived growth factor (PDGF), transforming growth factor-β (TGF-β), fibroblast growth factor (FGF) and other important molecules, participate in the process of angiogenesis [5]. Currently, with various antiangiogenic agents being approved for cancer treatment, antiangiogenic therapy for NSCLC has attracted increasing attention.
Bevacizumab is a humanized monoclonal antibody with high affinity to VEGF [6]. The VEGF family includes VEGF-A, VEGF-B, VEGF-C, VEGF-D and placental growth factor, which have been suggested to be overexpressed in NSCLC, providing necessary conditions for angiogenesis [7]. Bevacizumab has a high affinity and specificity for VEGF, and therefore, it can inhibit the growth of NSCLC [6], which provides a premise for it to become an agent for the treatment of NSCLC. The results of E4599, which is a randomized, controlled, multicenter clinical trial, show that bevacizumab combined with carboplatin and paclitaxel can significantly improve the response rate and prolong progression-free survival (PFS) and overall survival (OS) compared with carboplatin and paclitaxel alone for patients with unresectable and advanced non-squamous cell carcinoma who were chemotherapy naive [8]. Based on this finding, in 2005, the Food and Drug Administration approved bevacizumab combined with carboplatin and paclitaxel for the first-line treatment of advanced non-squamous cell carcinoma [9]. However, at present, only bevacizumab combined with cisplatin and paclitaxel has been approved for NSCLC treatment, and the differences in outcomes of bevacizumab combined with different chemotherapies have not been clearly determined.
Endostatin is a kind of natural protein that was first isolated and extracted by Judah Folkman in mouse tumor strains and showed the strong antiangiogenic and tumor growth inhibition effects [10]. Endostatin has a wide antitumor spectrum [11]. The specific antiangiogenic mechanism has not been fully elucidated at present. The main mechanism of the antitumor effect is that endostatin acts on the VEGF receptor KDR/Flk-1 and inhibits the signal transduction of VEGF, which therefore inhibits angiogenesis [12]. Endostar is a recombinant human endostatin with 9 added amino acids (MGGSHHHHH) [13] to maintain stability and a long half-life. A randomized, double-blind, controlled, multicenter phase III clinical trial from China showed that the combination of Endostar with vinorelbine and cisplatin can significantly improve the response rate, median time to progression (TTP), and quality of life of patients with advanced NSCLC compared with vinorelbine and cisplatin alone. There were no significant differences in adverse events, but its cardiotoxicity needs further attention [14]. Endostar was approved by the China Food and Drug Administration in 2005 for the treatment of NSCLC [15]. However, at present, only one phase 3 clinical trial has shown that the efficacy of Endostar combined chemotherapy, and the advantages of Endostar combined with other chemotherapeutics need more rigorous clinical trials for confirmation.
Currently, it is still not clear that antiangiogenic drugs combined with which chemotherapy regime can provide the most benefit to NSCLC patients. Besides, although the results from phase 3 clinical trials showed that NSCLC patients can benefit from bevacizumab combined with chemotherapy and Endostar combined with chemotherapy for first-line treatment, there is a lack of head-to-head clinical trials of the two schemes. Therefore, we used the Bayesian network meta-analysis method to indirectly compare the efficacy of bevacizumab combined with chemotherapy and Endostar combined with chemotherapy in the treatment of NSCLC. In addition, we used single-arm meta-analysis to synthesize published clinical trial results to comprehensively evaluate the therapeutic effect of bevacizumab combined with chemotherapy and Endostar combined with chemotherapy on NSCLC patients.
Inclusion and exclusion criteria were established for the purpose of our analysis. The inclusion criteria were as follows: the study type was a clinical trial or prospective study; the study population was NSCLC patients who had not received chemotherapy or NSCLC patients who had completed previous chemotherapy for more than or equal to 4 weeks; the interventions were bevacizumab combined with chemotherapy, Endostar combined with chemotherapy or chemotherapy alone; the study design involved studies with at least two arms and included at least the above two interventions or included the same intervention in the same study but used different chemotherapy agents; and studies that provided at least one of the following outcome measures: objective response rate (ORR), overall survival (OS), and progression free survival (PFS), studies that provided the event rate of adverse events after treatment were also included. The exclusion criteria included articles that could not be provided in full-text and non-English articles; and the following article types were also excluded: review articles, case reports, meeting abstracts, meta-analyses, cell animal experiments, etc. Three reviewers (YL, YY and AL) independently performed preliminary screening according to the titles and abstracts and then confirmed whether to include the studies by reading the full texts according to the inclusion and exclusion criteria. Any disputes regarding the included studies were resolved through discussion with the fourth reviewer (PL).

Data extraction
Three reviewers (YL, YY and AL) used a previously designed data extraction table to extract data independently. The extracted data included the general data of the study (such as the first author, study year, proportion of males, treatment plan, performance status, proportion of squamous cell carcinoma), the event rate of adverse events, the ORRs, hazard ratios (HRs) and 95% confidence intervals (CIs) of OS and PFS. If the HR and 95% CI were not provided in the study, we used Tierney's method to estimate them [16]. Any disputes or inconsistencies were discussed with the fourth reviewer (PL).

Quality assessment
Three reviewers (YL, YY and AL) independently evaluated the quality of the research methods of the included studies. The Cochrane Collaboration's risk of bias (ROB) tool was used to evaluate the quality of randomized controlled trials [17], and the Newcastle Ottawa Scale (NOS) was used to evaluate the quality of nonrandomized controlled trials and prospective studies [18]. For randomized controlled trials, if there were more than four "low risk" domains based on the ROB tool, the study was considered to have "high quality"; if there were two or more "low risk" domains, the study was considered to have "moderate quality"; and finally, if there were less than two "low risk" domains or more than one "high risk" domain, the study was considered to have "low quality". For nonrandomized controlled trials, if the total NOS score was 7-9, the study was considered to have "high quality"; if the total NOS score was less than 4, the study was considered to have "low quality"; and if the total NOS score was 4-6, the study was considered to have "moderate quality". Similarly, any disputes and inconsistencies were discussed with the fourth reviewer (PL). Statistical analysis R 3.5.1 software (The R Foundation, Vienna, Austria) was used for Bayesian meta-analysis. For the binary variable (ORR) and time-to-event data (HRs of OS and PFS), we used the "gemtc" package and "JAGS" package of R software, using the Markov chain Monte Carlo method [19] to simulate four different chains, with 40,000 iterations, 160,000 burn-ins and a thinning interval of 10. The Bayesian network meta-analysis was carried out so that each intervention could be compared indirectly, and the rank probabilities of various interventions can be obtained. For the comparison of ORR, the results are shown as odds ratios (ORs) and 95% credible intervals (CrIs); for the comparison of OS and PFS, the results are shown as HRs and 95% CrIs, and the results of all CrIs were bilateral. In addition, to ensure the credibility of the results, we used the "gemtc" package and "JAGS" package in R 3.5.1 software to analyze the ORR and HR of OS and PFS by nodesplit analysis to explore the consistency between direct comparisons and indirect comparisons. P > 0.05 indicates that the difference between the direct comparison and the indirect comparison was not statistically significant. P < 0.05 indicates that the difference between the direct comparison and the indirect comparison was statistically significant. Heterogeneity analysis was used to compare the degree of heterogeneity in the research results. When the P value was more than 0.1 and I 2 was less than 50%, the heterogeneity was not statistically significant. When the P value was less than 0.1 and I 2 was more than 50%, the heterogeneity was statistically significant.
Review Manager 5.3.4 software (Cochrane Library, Oxford, UK) and STATA 14.0 software (Stata Corp., College Station, TX) were used for single-arm meta-analysis. The ORR, OS, PFS, HRs of OS and PFS and the event rate of adverse events were used to obtain pooled results. The statistical model was selected according to the degree of heterogeneity. When the P value was more than 0.1 and I 2 was less than 50%, no statistically significant heterogeneity was indicated, so a fixed effects model was used. When the P value was less than 0.1 and I 2 was more than 50%, statistically significant heterogeneity was indicated, so a random effect models was used. At the same time, we used RevMan to evaluate publication bias intuitively through funnel plots and STATA to evaluate publication bias quantitatively through Begg's and Egger's tests. In addition, STATA was used to generate a network plot to describe the number of patients included in the study and the number of head-to-head comparisons.

Characteristics of the included studies
According to the search strategy, we retrieved 10,650 articles in PubMed, Web of Science and the Cochrane Library. Sun Yan et al. provided inadequate information [47] for our network meta-analysis, and therefore, we e-mailed the corresponding authors for further information. Finally, the author provided us with more detailed results of a phase 4 clinical trial of Endostar as additional sources for retrieval. Finally, a total of 29 articles were included in our study, including 4 nonrandomized controlled trials and 25 randomized controlled trials. The screening process of the articles is shown as a flow diagram in Fig. 1. The general characteristics of the patients in each study are shown in Table 1. The details of quality assessment are shown in Additional file 1: Figure S1.

Network meta-analysis of the outcome measures
To facilitate the study, we summarized the interventions involved in the articles as follows: platinum + gemcitabine + Endostar (Pla + Gem + Endo), p l a t i n u m + g e m c i t a b i n e + b e v a c z u m a b (Pla + Gem + Bev), platinum + gemcitabine (Pla + Gem), platinum + paclitaxel + Endostar (Pla + Pac + Endo), platinum + paclitaxel + bevacizumab (Pla + Pac + Bev), platinum + paclitaxel (Pla + Pac), platinum + pemetrexed + bevacizumab (Pla + Pem + Bev), platinum + pemetrexed (Pla + Pem), pemetrexed + bevacizumab (Pem + Bev), vinorelbine + platinum (Vin + Pla), vinorelbine + platinum + Endostar (Vin + Pla + Endo), gemcitabine + bevacizumab (Gem + Bev) and platinum + pemetrexed + Endostar (Pla + Pem + Endo). A total of 12,862 patients were treated with these interventions. The sample size of patients included in each intervention and the number of head-to-head comparisons were roughly described as a network plot, as shown in Fig. 2. The results of the Bayesian meta-analysis of ORR, OS and PFS are summarized in Fig. 3. The analyses of OS and PFS were not fully included for all interventions because some studies were unable to provide data related to ORR, OS or PFS.
For ORR, there were no significant differences (the CrI of the OR value included 1) in the improvement of the ORR of NSCLC patients between 4 different Endostar combined with chemotherapy regimens (Pla + Gem + Endo, Pla + Pac + Endo, and Vin + Pla + Endo, Pla + Pem + Endo) and 5 bevacizumab combined with chemotherapy regimens (Pla + Gem + Bev, Pla + Pac + Bev, Pla + Pem + Bev, Pem + Bev and Gem + Bev). There were no significant differences among the 4 Endostar combined with chemotherapy regimens. However, some Endostar combined with chemotherapy regimens had significant advantages over some chemotherapy regimens based on platinum. Pla + Gem + Endo was superior to Pla + Gem [OR: 0. 47  . Interestingly, for the improvement of ORR, Endostar combined with chemotherapy regimens (Pla + Gem + Endo, Pla + Pac + Endo, Vin + Pla + Endo and Pla + Pem + Endo) and bevacizumab combined with chemotherapy regimens (Pla + Gem + Bev, Pla + Pac + Bev, Pla + Pem + Bev and Pem + Bev), were significantly better than Pem alone, with ORs and 95% CrIs of 0. 17  For the time-to-event data of OS, there were no significant differences between the Endostar combined with chemotherapy regimens and the bevacizumab combined with chemotherapy regimens. For the bevacizumab combined with chemotherapy regimens, Pla + Pac + Bev was significantly better at prolonging OS than Pla + Gem [HR: 0.83 (0.71, 0.97)]; similarly, compared with Pla + Pac, Pla + Pac + Bev can also significantly reduce the HR [HR: The results of rank probabilities for ORR, OS and PFS were shown as heat maps in Fig. 4 (See more detailed data in Additional file 1: Table S1). For the time-to-event data of PFS, there were no significant differences between the Endostar combined with chemotherapy regimens and the bevacizumab combined with chemotherapy regimens, similarly. And for the bevacizumab combined with chemotherapy regimens, Pla + Pac + Bev was superior to Pla + Gem [HR: 0.57 (0.42, 0.79)] and Pla + Pac [HR: 1.9 (1.5, 2.4)] (Pla + Pac vs Pla + Pac + Bev).
According to the results of network meta-analysis for ORR, Pla + Pem + Bev was the most likely best intervention (0.288) and Pla + Pac + Bev (0.255) was the most likely second best intervention. In terms of the time-to-event data of OS, Pla + Pem + Endo was the most likely best intervention (0.423), Pem + Bev (0.219) was the most likely second best intervention. And for the time-to-event data of PFS, the most likely best intervention was Pla + Gem + Endo (0.302), and the most likely second best intervention was Pla + Pac + Bev (0.340).

Nodesplit and heterogeneity analyses
The results of nodesplit analysis and heterogeneity were shown in Fig. 5. To test the consistency and heterogeneity

Single-arm meta-analysis
The forest plots of single-arm meta-analysis are shown in Fig. 6 (ORR) and Fig. 7 (OS and PFS). The forest plots of the event rate of adverse events are shown in Additional file 1: Figure S2.             Figure S3.

Discussion
Angiogenesis plays an important role in tumorigenesis and development and is related to tumor proliferation, invasion and metastasis [4]. The results of clinical studies has suggested the combination of antiangiogenic agents and chemotherapy was effective, which can improve the ORR and prolong the OS and PFS for NSCLC patients. However, bevacizumab combined with chemotherapy and Endostar combined with chemotherapy are still lack of head-to-head clinical trials, which limits the clinical application of antiangiogenic agents combined with chemotherapy in the treatment of NSCLC. Therefore, by conducting the current Bayesian network meta-analysis, the effects of 14 interventions included in our study on improving ORR, prolonging PFS and OS were compared indirectly, so as to provide possible suggestions for further clinical trials and rational application of antiangiogenic agents and chemotherapy in the treatment of NSCLC. Our study showed that antiangiogenic agents combined with dual chemotherapy agents containing platinum may have a better effect on NSCLC patients. In terms of ORR, bevacizumab combined with chemotheray may have a better effect for the treatment of NSCLC; and in terms of OS and PFS, Endostar combined with chemotherapy may be superior to bevacizumab combined with chemotherapy.
The results of the single-arm meta-analysis showed that the combination of antiangiogenic agents and chemotherapy agents could improve ORR [pooled ORR and 95% CI: 0. 35  , which showed that antiangiogenic agents combined with chemotherapy was more effective than chemotherapy alone. However, at present, it is still unclear that antiangiogenic agents in combination with which chemotherapy regimen can provide more benefits to patients in the treatment of NSCLC. Based on the first-ranked and second-ranked interventions of rank probabilities, our analysis found that, in terms of ORR, antiangiogenic agents combined with Pla + Pem or Pla + Pac may have better therapeutic effects; in terms of OS, antiangiogenic agents combined with Pla + Pem or Pem may be superior to prolonging OS and in terms of PFS, combined with Pla + Gem or Pla + Pac may have better effects. Therefore, our results showed than antiangiogenic agents combined with Pla + Pem, Pla + Pac or Pla + Pem may benefit patients more in the treatment of NSCLC. Among them, antiangiogenic agents combined with Pla + Pem seems to be more conducive to improving ORR and prolonging OS and combined with Pla + Pac seems to be the first choice for prolonging OS and PFS.
The combinations of Endostar, an antiangiogenic agent, combined with chemotherapy and bevacizumab combined with chemotherapy have shown good results. However, the comparison between Endostar combined with chemotherapy and bevacizumab combined with chemotherapy lacks head-to-head clinical trials. Our results suggested that there was no significant difference in the effect of improving ORR and prolonging OS and PFS in NSCLC patients between Endostar combined with chemotherapy and bevacizumab combined with chemotherapy. Although there were no significant differences between these interventions, the rank positions indicated that Pla + Pem + Bev and Pla + Pac + Bev were in a more advanced position for improving ORR, and in terms of OS, Pla + Gem + Endo may be the best choice among these interventions. And interesting, for prolnging PFS, our results showed that Pla + Pem + Endo may be the first-ranked regimen, however, as the only bevacizumab combined with chemotherapy regimen for the treatment of non-squamous NSCLC aproved by FDA, Pla + Pac + Bev was the second-ranked regimen for improving PFS. Our analysis indicaded that bevacizumab combined with chemotherapy seems more effective for improving ORR and Endostar seems more effective for prolonging OS and PFS.
For the comparison of the different Endostar combined with chemotherapy regimens, a 4-phase clinical study showed that there were no significant differences in improving the ORR, OS and TTP (time to progression) among Vin + Pla + Endo, Pla + Gem + Endo and Pla + Pac + Endo [47]. However, for Pla + Pem + Endo, there was few head-to-head comparisons with other Endostar combined chemotherapy regimens. Our network meta-analysis showed that compared with Vin + Pla + Endo, Pla + Gem + Endo and Pla + Pac + Endo, there were no statistical difference in improving ORR and prolonging OS and PFS. However, the results of single-arm meta-analysis showed that some differences still existed among them. Compared with Pla + Gem + Endo, the ORR [ 0. 22 0.06, 0.49)]. The reason may be related to the differences and inconsistencies between the studies and the limited number of studies. In the future, head-to-head randomized controlled trials are needed to determine the relationship between the above strategies for improving the prognosis of patients to provide more reasonable treatments for NSCLC patients.
For adverse events, the event rate of leukopenia [0.57 (0.32, 0.82)] was the highest among the four most common adverse events between Endostar combined with chemotherapy and bevacizumab combined with chemotherapy. It is worth noting that the common adverse events of bevacizumab combined with chemotherapy, such as hypertension, proteinuria and thromboembolism, have not been studied in the study of Endostar combined with chemotherapy. Compared with bevacizumab combined with chemotherapy, in Pla + Pac + Endo, the event rate of leukopenia was the highest [0.87 (0.79, 0.95)], while the incidence of vomiting was the second lowest [0.15 (0.06, 0.23)]. In addition, in Pla + Pem + Endo, the event rate of vomiting was the highest [0.86 (0.77, 0.95)] and the incidence of thrombocytopenia was the second lowest [0.14 (0.05, 0.23)]. Gem + Bev had a low incidence in both anemia [0.09 (0.01, 0.18)] and thrombocytopenia [0.12 (0.02, 0.21)]. However, due to the high heterogeneity, limited number and potential publication bias of studies, further studies are needed to confirm these conclusions.
For our network meta-analysis, there were still some limitations. First, we included 11 moderate-quality studies and 18 high-quality studies: of the 25 randomized controlled trials, 10 were moderate-quality articles and 15 were high-quality articles; the other 4 were nonrandomized controlled trials and prospective studies, of which 3 were high-quality articles and 1 was a moderate quality article according to the NOS. Although the number of included nonrandomized controlled trials and prospective studies was relatively small, the methodological design of the randomized clinical trials was still more reliable than that of the nonrandomized controlled trials, which may lead to inconsistency in our results. Second, for the patients who participated in the study, squamous cell carcinoma and non-squamous cell carcinoma were not distinguished, which may have some impact on the consistency of our results. For squamous cell carcinoma, there were no independent study data and thus, we failed to observe the curative effect of Endostar combined with chemotherapy and bevacizumab combined with chemotherapy in squamous cell carcinoma. Third, in order to facilitate the analysis, we did not make a strict distinction between the dosage and the method of administration. In addition, original data and laboratory data was lacking in the current meta-analysis. Finally, our results were based on a Bayesian network meta-analysis and the statistical analysis of various interventions; however, the performance of various interventions in real patients still needs to be confirmed by head-to-head clinical trials.

Conclusions
In summary, the combination of antiangiogenic agents with platinum-containing dual drug chemotherapy can improve NSCLC patients' benefit. The combination of Endostar and platinum-containing dual drugs may be a better choice to prolong OS and PFS. More clincial trials are needed to ensure the reasonable use of antiangiogenic combined with chemotheray regimens for NSCLC patients in the future.