Systemic therapy as First-Line Treatment for Patients with Metastatic Colorectal Cancer: A Systematic Review and Network meta-analysis of Randomized Clinical Trials

Purpose: To assess the relative ecacy and safety of rst-line systemic therapies in patients with metastatic CRC (mCRC). Experimental Design: A comprehensive literature review was conducted including MEDLINE, Embase, and the Cochrane Central Registry of Controlled Trials for phase II or III Randomized Controlled Trials (RCTs) published up to and including July 15, 2019.We included RCTs in which at least 1 intervention was either a chemotherapeutic agents (such as, uorouracil, irinotecan, or oxaliplatin) or antibodies targeting angiogenesis (such as, bevacizumab) or agents that act on the epidermal growth factor receptor pathway (such as, cetuximab and panitumumab) or studies reported at least one of the following outcomes: Overall Survival (OS), Progression-Free Survival (PFS) and/or serious adverse events (SAEs). Using a random effect model, we performed a Bayesian network meta-analysis to analyze the probability of optimal therapeutic regime obtained from direct comparisons with indirect evidences. We estimated hazard ratios (HRs) for OS and PFS. Results: A total of 30 RCTs comprising (12, 146 patients with mCRC comparing 25 different strategies were included. The triple combination FOLFOXIRI [uorouracil, leucovorin, oxaliplatin, and irinotecan] plus bevacizumab provided signicant survival benets with improved OS over all other treatments. The network meta-analysis also indicated a signicant advantage of using FOLFOXIRI plus bevacizumab in comparison to other treatment strategies for PFS. Besides, FOLFOXIRI plus bevacizumab was associated with the lowest risk of serious adverse events. Conclusions: Our study supported the use of FOLFOXIRI plus bevacizumab as the best rst-line regimen and potentially effective and safe strategy for the management of patients with mCRC.


Introduction
Colorectal cancer (CRC) ranks third amongst all malignant neoplasms and continues to be the leading cause of cancer-associated mortality, worldwide (Hegde et al., 2008). Approximately, 25% of patients with CRC present with liver metastasis at the initial diagnosis or will develop liver metastasis during the course of their disease (Chow and Chok, 2019). In spite of the emergence of highly effective chemotherapy and advances in surgical techniques, the pool of patients with liver and/or lung-isolated metastasis has expanded and for the majority of patients with metastatic CRC (mCRC), the treatment remains a clinical challenge (Park et al., 2013). Indeed, for many years 5-uorouracil (FU)-based regimens has been the backbone of systemic therapy for mCRC. Recent incremental advancements in the systemic therapy for mCRC have been signi cantly facilitated with the introduction of several new cytotoxic and biologic agents.
Systemic therapy includes combinations of chemotherapeutic agents (oxaliplatin, irinotecan or uorouracil) alone or in combination with monoclonal antibodies targeting epidermal growth factor receptor (EGFR; cetuximab and panitumumab) or vascular endothelial growth factor receptor (VEGFR; bevacizumab), thereby providing distinctly effective rst-line therapeutic regimens for mCRC (Saltz et al., 2004). However, head-to-head randomized trials comparing these therapeutic regimens mentioned above are still lacking, thus there is no evidence to guide optimal regimen for patients mCRC. To overcome these limitations, using a network meta-analysis (NMA) approach, we compared and evaluated the relative therapeutic e cacies of all possible combinations of treatments, by simultaneous integration of direct evidence from head-to-head trials and indirect evidence to rank the different treatments for mCRC.

Literature Search
Literature screening was performed according to the method outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Green, 2011). Institutional review board approval was not required.
We conducted a comprehensive literature search of electronic databases including MEDLINE, EMBASE, and the Cochrane Central Register of Controlled Trials (Central) databases from inception up to and including July 15, 2019. A combination of MeSH-terms and keywords strategy was applied as follows: "Advance or metastatic colon cancer", "hepatic metastases, liver metastases" "immunotherapy and targeted therapy" (sTable1 in supplement). Also, the references of the selected articles and reviews were manually retrieved to obtain all potentially relevant studies. Retrieved articles were screened and reviewed for their eligibility by two independent reviewers (SX, YBE). Differences in the determination of the study's eligibility were resolved by consensus or through discussion with a third adjudicator (AS). The language of publication was restricted to English.

Study Selection
We included phase II or III randomized controlled trials (RCT) that met the following inclusion criteria: (a) the study subjects were patients with mCRC; (b) at least 1 of the interventions compared in the trial was either a chemotherapeutic agents (such as, uorouracil, irinotecan, or oxaliplatin) or antibodies targeting angiogenesis (such as, bevacizumab) or agents that act on the EGFR-related pathway (such as, cetuximab and panitumumab); (c) the primary outcome was overall survival (OS), progression-free survival (PFS), and/or serious adverse events (SAEs) of greater than or equal to grade 3 according to the National Cancer Institute's (NCI) Common Terminology Criteria for Adverse Events (CTCAE). We excluded studies that were not RCTs and had unavailable data.

Data Extraction and Risk of Bias Assessment
The data extraction from each included study was performed independently by two reviewers (SX, AS) and entered into a standardized, predesigned Microsoft Excel form. The following data were extracted: the rst author, the year of publication, country, patient characteristics, treatment strategies, sample size, number of patients evaluated for response, dose, and schedule, median cycles received, and outcomes (median OS and median PFS). For PFS and OS, we extracted the hazard ratio (HR) with a 95% con dence interval (95% CI) if available. However, when HRs and corresponding CIs were not reported, we estimated them by reconstructing individual patient data from published Kaplan-Meier curves with methods described by Guyot and colleagues(Guyot et al., 2012). Authors of included studies were contacted if important data were unclear or not reported. The risk of bias in randomized trials was assessed independently by the reviewers (SX and YBE) using the Cochrane Collaboration tool and the risk-of-bias (RoB 2.0) tool (Higgins and Altman, 2008). Any disagreements were resolved through consensus.

Data Synthesis and Analysis
We synthesized evidence for 3 outcomes: PFS, OS and any SAEs. For each outcome, we performed a Bayesian NMA using a Markov Chain Monte Carlo (MCMC) simulation technique with 100 000 iterations in each of the 3 chains. Noninformative priors (ie, N[0, 10000]) were chosen for the effect parameters. The analysis was performed under the xed-effect model, as only 1 trial provided direct evidence for most of the treatment comparisons. However, a random-effects (RE) model was also performed as sensitivity analysis and model ts were compared using deviance information criteria (DIC) (Higgins, 2013). In the comparison of any 2 models, we deemed a better t model if its DIC was less than the DIC of the other model by at least 5. Heterogeneity in the network was assessed with the Cochrane Q (χ2) test and quanti ed using the I2 statistic within each pairwise comparison when 2 or more trials were available for the comparison (Higgins and Thompson, 2002). Because it was uncommon to have both direct and indirect evidence for most comparisons in our networks, we assumed coherence for our analysis (ie, direct and indirect evidence, when both available for a given comparison, were statistically similar). To test the robustness of this assumption we used the node-splitting method to assess whether there was incoherence in any closed loops (Higgins, 2013;Higgins and Thompson, 2002). Relative effects of treatments are reported as HR for survival outcomes (PFS and OS) and odds ratio (OR) for binary outcomes (SAEs) along with corresponding 95% credible intervals (CrIs), the Bayesian equivalent of 95% CIs. We estimated the overall ranks of treatments by calculating the surface under the cumulative ranking curve (SUCRA) for each (Salanti et al., 2011). The SUCRA index ranges between 0 (or 0%) and 1 (or 100%), where the treatments with highest and lowest SUCRA are considered to be the best and worst treatments, respectively. Network meta-analysis was performed in WinBUGS software (version 1.4.3, MRC Biostatistics Unit) interfacing through R software.

Overall characteristics of selected studies and Quality of Evidence
The PRISMA owchart of included studies is presented in sFigure.1 (shown in supplement). After the exclusion of duplicate studies, a total of 557 records was initially identi ed through our literature search.
After a detailed assessment by the full-text review, 30 trials comprising 12,146 patients with mCRC were included in this meta-analysis ( Figure 1). The characteristics of 30 RCTs included in the meta-analysis are summarized in Table 1 The study sample sizes ranged from 25 to 599. These studies were published between 2005 and 2019. The risk of bias and the quality assessment in all studies were presented in supplementary table (sTable 20), which indicated that the quality of the included studies was reliable.
Moreover, according to the MCMC model, I 2 was estimated to be 0.00%. Therefore, there is no heterogeneity in the data, and the results of NMA are stable and reliable.

Overall Survival
Twenty-ve trials comprising 11,175 patients with MCRC comparing 21 treatments were included in the OS analysis (supplementary gure, sFigure 2A). The results indicated that the FOLFOXIRI/Bev treatment strategy was associated with improved OS bene ts compared with all other treatments. The key comparison treatments included FOLFOXIRI/Bev vs FOLFOX/Bev with HR, 1.03 (95% CrI, 0.69-1.52), and FOLFOXIRI/Bev vs FOLFIRI/Bev with HR, 1.07 (95% CrI, 0.84-1.34). The estimated SUCRA values were 77.2 and 77.1% for FOLFOXIRI/Bev and FOLFOX/Bev treatment strategies, respectively, suggesting that these two treatment strategies exhibited the highest probability of being the best treatment for improving OS of patients with mCRC (sTable 2A in the Supplement).

Progression-Free Survival
Twenty-two trials comprising 9,588 patients with mCRC, comparing 25 treatments were included in the PFS analysis (supplementary gure sFigure 2B). FOLFOXIRI/Bev treatment strategy exhibited a signi cantly higher PFS compared with other strategies. Consistently, the SUCRA analysis also suggested that FOLFOXIRI/Bev treatment strategy was the best treatment in promoting PFS (SUCRA: 93.2%), followed by FOLFOXIRI (SUCRA: 79.9%), whereas CapeOX/Cetuximab treatment strategy was least likely to be the optimal treatment strategy in improving PFS (SUCRA: 17.2%) (Supplementary table, sTable 2B).

Serious Adverse Events
Eighteen trials comprising 8,986 patients with mCRC, comparing 19 treatment strategies reported adverse events of grade 3 or higher (supplementary gure, sFigure 2C). FOLFOXIRI/Bev treatment strategy was signi cantly associated with a lower risk of SAEs compared with all other treatments. Consistently, the SUCRA analysis also suggested that FOLFOXIRI/Bev and XELIRI/Bev treatment strategies exhibited the lowest risk of SAEs with SUCRA values of 96.3% and 83%, respectively. Besides, FOLFOX/Bev/Onartuzumab treatment strategy was associated with a higher risk of SAEs compared with all other treatments (SUCRA values was 4.6%) (sTable 2C in the Supplement).

Discussion
Incremental advancements have been made in mCRC therapy ever since the introduction of 5-FU over 40 years ago (Hammond et al., 2016). Moreover, the treatment of mCRC has been facilitated signi cantly with the introduction of several new cytotoxic and biologic agents to the 5-FU regimen. Notably, combination regimens that incorporate infusional schedules of 5-FU in various combinations, including XELOX regimen (oxaliplatin and capecitabine), FOLFOX regimen (leucovorin, 5-FU and oxaliplatin) and FOLFIRI regimen (leucovorin, 5-FU and irinotecan), with or without monoclonal antibody, have signi cantly improved the clinical outcomes and median overall survival of patients with mCRC.
In this systematic review and NMA meta-analysis, we estimated the relative e cacy of the different combinations of treatment strategies for outcomes involving OS and PFS in patients with mCRC. Overall survival remains the fundamental endpoint in clinical trials; this meta-analysis found that triple combination FOLFOXIRI [ uorouracil, leucovorin, oxaliplatin, and irinotecan] plus bevacizumab provided signi cant survival bene ts over all the other treatments which were in accord with recent studies (Cremolini et al., 2015). Consistently, a meta-analysis also showed a signi cant increase in OS with FOLFOXIRI plus bevacizumab in patients with mCRC (Tomasello et al., 2017). Notably, FOLFOXIRI plus bevacizumab was also found to be most effective in promoting PFS. These results are also consistent with the TRIBE study, which indicated signi cant OS and PFS bene t with the addition of oxaliplatin to FOLFIRI plus bevacizumab in patients with mCRC (Cremolini et al., 2015). Other recent trials and studies (VISNU-1 trial and QUATTRO Study) have also shown that mCRC patients who received FOLFOXIRI plus bevacizumab exhibited statistically signi cant improved PFS and OS (Oki et al., 2018;Sastre et al., 2019). Survival bene ts needed to be justi ed against the toxicity of chemotherapy. In the majority of mCRC patients with advanced, systemic treatment remains noncurative, and thus the quality of life becomes a priority. In this meta-analysis, rates of SAEs were high for all treatment strategies except for FOLFOXIRI plus bevacizumab regimen, which exhibited no signi cant increase in the adverse events compared with other treatments.
Using appropriate statistical approaches, this network meta-analysis is the rst study to provide an indirect comparison for the currently available systemic therapy for patients with mCRC, based on all available information from phase II and III randomized trials. This NMA is acknowledged to have several limitations. First, some included studies lacked su cient comparisons, which may have a certain impact on the result. In addition, the collected results from the included studies were uneven and the sample size of few studies on some drugs was relatively small.
Overall, our study supported the use of FOLFOXIRI-bevacizumab as the best rst-line regimen and potentially effective and safe strategy for the management of patients with mCRC. Furthermore, our upto-date analysis provides new insights into existing controversies on systemic therapy for patients with mCRC.