A Systematic Review and Network Meta-Analysis of Biologic Agents in the First Line Setting for Advanced Colorectal Cancer

Background Epithelial growth factor receptor inhibitors (EGFRis) and bevacizumab (BEV) are used in combination with chemotherapy for the treatment of metastatic colorectal cancer (mCRC). However, few randomized controlled trials (RCTs) have directly compared their relative efficacy on progression-free survival (PFS) and overall survival (OS). Methods We conducted a systematic review of first-line RCTs comparing (1) EGFRis vs. BEV, with chemotherapy in both arms (2) EGFRis + chemotherapy vs. chemotherapy alone, or (3) BEV + chemotherapy vs. chemotherapy alone, using Cochrane methodology. Data on and PFS and OS were extracted using the Parmar method. Pairwise meta-analyses and Bayesian network meta-analyses (NMA) were conducted to estimate the direct, indirect and combined PFS and OS hazard ratios (HRs) comparing EGFRis to BEV. Results Seventeen RCTs contained extractable data for quantitative analysis. Combining direct and indirect data using an NMA did not show a statistical difference between EGFRis versus BEV (PFS HR = 1.11 (95% CR: 0.92–1.36) and OS HR = 0.91 (95% CR: 0.75–1.09)). Direct meta-analysis (3 RCTs), indirect (14 RCTs) and combined (17 RCTs) NMA of PFS HRs were concordant and did not show a difference between EGFRis and BEV. Meta-analysis of OS using direct evidence, largely influenced by one trial, showed an improvement with EGFRis therapy (HR = 0.79 (95% CR: 0.65–0.98)), while indirect and combined NMA of OS did not show a difference between EGFRis and BEV Successive inclusions of trials over time in the combined NMA did not show superiority of EGFRis over BEV. Conclusions Our findings did not support OS or PFS benefits of EGFRis over BEV in first-line mCRC.

, metastatic pancreatic cancer [27], adjuvant treatment for pancreatic cancer [28] and Hodgkins lymphoma [29] to simultaneously examine the relative efficacy of multiple treatments by synthesizing both direct and indirect evidence.
In this study, we carried out a systematic review of all randomized controlled trials (RCTs) that compared systemic chemotherapy regimens with and without bevacizumab, cetuximab or panitumumab for the first-line treatment of mCRC and conducted both direct meta-analysis, and indirect and combined NMAs to assess the impact of these agents on PFS and OS.

Search strategy and selection criteria
We performed a systematic review to identify studies which examined the survival outcomes of mCRC patients. Under consideration for inclusion were all RCTs which compared chemotherapy treatment alone to either i) chemotherapy treatment in combination with BEV or ii) chemotherapy in combination with EGFRis. RCTs that directly compared chemotherapy treatment with EGFRis against chemotherapy combined with BEV were also considered. Studies were included regardless of the chemotherapy backbone as long as the backbone was identical in both the treatment and control arms. Studies were included only if the patients were being treated with first-line therapy for the treatment of mCRC. Those studies with more than two treatment arms were included if one of more of the arms included an eligible comparison; only arms addressing eligible comparisons were included in the analysis. For EGFRis trials, only data from K-RAS participants with wild-type K-RAS Exon 2 tumors was included.
We excluded non-randomized trials, as well as trials involving non-metastatic colorectal cancer patients. Trials that included radiotherapy, hormonal therapy, gene therapy or other immunologic therapy in one of the arms were excluded. Studies with a comparison of chemotherapy and a VEGFi or an EGFRi against no treatment (best supportive care), or against a VEGFi or EGFRi alone were excluded. When several reports discussed the same trial, the report with the most recent data was included.
We searched Medline, Embase, and the Central Registry of Controlled Trials of the Cochrane Library. All databases were updated through to the second week of September, 2014. We did not place any language restrictions on the search. The complete search strategy employed has been provided (S1 Text). Our review has been reported using the PRISMA reporting guidelines (S1 PRISMA Checklist).

Data extraction
Data was extracted by two independent reviewers with discrepancies between the reviewers discussed prior to selecting trials for inclusion in the systematic review and prior to inclusion in the meta-analysis. Unresolved discrepencies were reviewed by a third reviewer. We recorded primary author, trial ID, treatment comparison, primary and secondary outcomes studied, location of the trial, recruitment period, number of patients randomized and evaluated in each treatment arm, age, gender, and K-RAS status.
The studies were organized into three treatment groups; chemotherapy with BEV against chemotherapy alone, chemotherapy with an EGFRi against chemotherapy alone, and chemotherapy with BEV against chemotherapy with an EGFRi. For our analysis, studies examining cetuximab or panitumumab were grouped together as EGFRi. Studies with different chemotherapy backbones, methods of chemotherapy administration, or methods of antibody therapy administration were grouped according to the type of antibody therapy used. In trials involving a comparison with an EGFRi, only data pertaining to wild-type K-RAS Exon 2 patients was extracted.
For each trial, we recorded PFS and/or OS of the treatment and control arms, the hazard ratio (HR), log-rank p-value, and confidence intervals when available. In studies where the HR, p-value, or confidence intervals were not provided, it was calculated from the log-rank p value, the number of events in each arm and the number of randomized patients in each arm using the Parmar method [30].

Statistical analysis
Pairwise meta-analyses were conducted to examine treatment regimens that were directly compared in the studies; chemotherapy vs chemotherapy with EGFRis, chemotherapy vs chemotherapy with BEV, and EGFRis vs BEV with chemotherapy in both arms. The results were combined into forest plots using Review Manager, version 5.2, using the random-effects model. To assess the comparability of included studies, between-study heterogeneity was estimated and reported using the I 2 statistic; the value of I 2 lies between 0% and 100%, where 0% indicates no observed heterogeneity and larger values show increasing heterogeneity [31].
We conducted a Bayesian NMA to examine the indirect comparison of EGFRis vs. BEV through the intermediate treatment of chemotherapy alone, and to combine the indirect comparison with the direct comparison, using WinBUGS, version 1.4.3. A detailed explanation of the statistical method [32] employed for the NMA has been provided (S2 Text). Bayesian NMAs were performed at three time points, which included: i) trials published prior to the FIRE-3 trial, [19] ii) trials up to and including FIRE-3, and iii) all trials published up to and including the CALGB 80405 trial [20], which was the last trial found by the literature search. In addition, a Bayesian NMA was conducted for all trials excluding FIRE-3. Sensitivity analyses were conducted to explore the effect of adjusting for the types of chemotherapy backbone (oxaliplatin-based, irinotecan-based or fluoropyrimidines alone) and the mode of fluoropyrimidines delivery (bolus or infusional). This was performed by including those effects as covariates in the meta-regression of the Bayesian NMA.
The results were presented according to the guidelines of the Quality of Reporting of Metaanalyses (QUOROM) and International Society for Pharmacoeconomics and Outcomes Research (ISPOR) [33,34].

Literature Search Results
Our electronic search of Medline, Embase and the Cochrane Central Register of Controlled Trials databases yielded 2435 potentially relevant articles. Our manual search through the 2013 and 2014 ASCO General Meeting abstracts produced an additional 62 results. Following a deletion of duplicate results from different databases, there were 1581 records. Ultimately, we identified 17 unique studies for inclusion in the meta-analysis, including 2 ASCO abstracts (Fig 1).

Study Quality
The quality of studies included in the NMA was evaluated using the Cochrane risk of bias tool [35]. A list of biases was summarized (S1 and S2 Figs). Publication bias was evaluated by constructing funnel plots for the studies included in each direct comparison for OS (S3 Fig). The three plots for OS were all symmetrical and show no evidence of publication bias. All included studies were randomized and followed intention-to-treat analysis for the primary endpoints (PFS and OS). Two of the studies (CALGB 80405, ITACA) were published only in abstract form, and we were unable to judge whether selection, allocation, assessment, attrition and/or reporting bias were present. All of the trials reported median PFS, and all except for one (Kabbinavar et al, 2003), provided extractable data for OS. Nine of the studies did not blind the patients and assessors, leading to a possible assessment bias with regards to our primary endpoint, PFS. Heterogeneity was present in some pairwise treatment comparisons based on I 2 ; however, the studies were comparable in terms of patient characteristics and outcomes.

Trial Characteristics
All included studies were randomized and specific for the first-line treatment of metastatic colorectal cancer. Table 1 summarizes the baseline characteristics of the included trials. Each trial had a chemotherapy regimen in both arms, and either BEV or EGFRi in at least one arm of the trial. In total, BEV plus chemotherapy was compared to chemotherapy alone in eight studies, an EGFRi plus chemotherapy was compared to chemotherapy alone in six studies, and BEV was compared to an EGFRi-with chemotherapy administered in both arms-in three studies. All trials included in the meta-analysis reported median PFS and OS.
All of the included studies were comparable in terms of patient characteristics. The PFS of the chemotherapy-only reference arms ranged between 5.2-8.7 months, while the OS of these reference arms ranged between 13.8-22.0 months. Median PFS and OS were lower in the chemotherapy reference arm in trials examining the efficacy of BEV than in trials examining an EGFRi. Of the fourteen trials that compared an EGFRi or BEV with chemotherapy alone, eleven found a statistically significant difference in PFS, and five found a statistically significant difference in OS. Three studies comparing BEV and EGFRis directly did not find statistically significant differences in PFS, and one study found a difference in OS.

Comparison of regimens: Pairwise direct meta-analyses
Pairwise comparisons of trials examining the efficacy of the same antibody therapy were made first using a random-effects model. Direct pairwise meta-analyses comparing EGFRis versus BEV with chemotherapy in both arms did not detect a difference between the two arms with respect to PFS, HR = 1.02 (CI: 0.93-1.13). However, with respect to OS, a statistically significant difference was seen in favor of the EGFRis arm, HR = 0.79 (CI: 0.65-0.98). The results of PFS and OS comparisons are shown in Figs 3 and 4, respectively. Forest plots of the hazard ratios for PFS and OS between EGFRis vs. chemotherapy alone, and between BEV and chemotherapy alone are available (S4 and S5 Figs, respectively).

Sensitivity Analysis
The results of sensitivity analyses adjusting for the effect of the types of chemotherapy backbone and mode of fluoropyrimidine delivery (bolus vs. infusional) have been provided (S7 Fig). The findings are the same as the main base case results of the network meta-analyses for OS and PFS.

Discussion
We conducted a systematic review and NMA of randomized controlled trials to examine whether EGFRis or BEV is associated with improved PFS and OS in the first-line treatment of patients with mCRC. The trials included in our study compared either EGFRis with chemotherapy vs. chemotherapy alone, BEV with chemotherapy vs. chemotherapy alone, or EGFRis vs. BEV with chemotherapy in both arms. Pairwise meta-analyses were conducted to compare direct evidence, and an NMA was conducted using the intermediate treatment of chemotherapy for an indirect comparison. The direct and indirect evidence was combined to determine which therapy better improves survival outcomes. The results of our NMA did not show a benefit to OS with EGFRi therapy, a trend which was also seen by the indirect comparison of treatment regimens. However, results from direct meta-analysis showed an improvement in survival with EGFRi therapy. This may be due to the large weight of the FIRE-3 trial (592 patients) on our direct analysis, which found a statistically significant improvement in OS with EGFRi therapy (HR = 0.77, p = 0.017). The results of the larger direct comparison CALGB 80405 trial (1137 patients)-the only one powered for OSwere congruent with our NMA results. Examining the OS HR with and without the inclusion of FIRE-3 results demonstrates the impact of the FIRE-3 trial on our combined results (HR = 0.91 (95% CR: 0.75-1.09) and HR = 0.94 (95% CR: 0.76-1.15)), respectively. In the FIRE-3 trial, the separation of the Kaplan-Meyer OS curves only after 18 months of treatment, which combined with the lack of differences in the PFS, suggests the improvement in the OS may be related to post-progression events. This phenomenon, which FIRE-3 investigators suggested may be related to the increased depth of response to EGFRi therapy, was not observed in other trials comparing EGFRi vs. chemotherapy alone [36], CALGB 80405, our indirect comparisons of EGFRis vs. BEV, or our combined NMAs. The discrepancy about the OS Two separate sensitivity analyses were conducted to adjust for the possible confounding effect of 1) chemotherapy backbone and 2) mode of fluoropyrimidine delivery (bolus vs. infusional) on our results. PFS and OS have been shown in the literature [37,38]-as well as in the control arms of trials included in our study (Table 1)-to vary with the chemotherapy regimen used, and were adjusted for. Similarly, we adjusted for the potential confounder of fluoropyrimidine delivery mode (bolus vs. infusional) [39]. Our sensitivity analyses showed similar results between each sensitivity analysis, and the unadjusted NMA, for both PFS and OS, suggesting the robustness of our results.
The results of our pairwise meta-analyses showed an improvement in survival with the addition of EGFRis or BEV to chemotherapy, which is consistent with similar studies in the literature and previous pairwise meta-analyses [40,41,42,43]. The heterogeneity in our study may be due to the different chemotherapy backbones used in the studies, as variation in survival outcomes was also observed in the chemotherapy reference arms. Furthermore, even though the funnel plots were symmetrical and did not suggest evidence of publication bias, their ability to detect publication bias was less sensitive due to the relatively small number of trials in each forest plot [44]. The EGFRi pairwise meta-analysis also included studies with different biologics added (cetuximab and panitumumab), which may have contributed to heterogeneity. Studies included in the NMA were comparable in terms of patient characteristics.
In our analysis, we examine K-RAS wild type tumors because K-RAS was previously recognized and accepted as the biomarker of choice for selecting patients for EGFR inhibitors prior to 2014. Recently, pan RAS (as known as extended RAS) has been recognized as the biomarker of choice for selecting patients for EGFRis since 2014 [45]. Therefore, in future analyses, it would be important to look at how the BEV vs. EGFRi therapies compare in terms of outcomes with pan RAS wide type (i.e. also with exon 3 and NRAS wild-type patients) [44]. Updated data from FIRE-3, PEAK, and CALGB 80405 trials will also provide further evidence to refine these findings.

Conclusion
Our NMA reviewed and analyzed the existing literature for RCTs examining EGFRi and BEV treatments for metastatic colorectal cancer in the first-line setting. The results of our NMA did not show a statistical difference between EGFRis and BEV therapies with regards to both PFS and OS. The results of the NMA were congruent with indirect analysis with respect to both PFS and OS, as well as with direct analysis with respect to PFS. The findings of CALGB appeared to be congruent with the collective synthesis of the existing literature, while the findings of FIRE-3 appeared to be incongruent with the remaining literature. Further evidence from ongoing trials, which directly compare EGFRis and BEV therapies, will further validate of our results.