The Prognostic Role of BRAF Mutation in Metastatic Colorectal Cancer Receiving Anti-EGFR Monoclonal Antibodies: A Meta-Analysis

Background BRAF mutation has been investigated as a prognostic factor in metastatic colorectal cancer (mCRC) undergoing anti-EGFR monoclonal antibodies (moAbs), but current results are still inconclusive. The aim of this meta-analysis was to evaluate the relationship between BRAF mutation status and the prognosis of mCRC patients treated with moAbs. Methods Eligible studies were identified by systematically searching Pubmed, the Cochrane Library, Web of Knowledge, and OVID. Risk ratio (RR) for overall response rate (ORR), Hazard ratios (HRs) for Progression free survival (PFS) and Overall survival (OS) were extracted or calculated. Prespecified subgroup analyses were conducted in KRAS wild-type and in different study types. The source of between-trial variation was explored by sensitivity analyses. Quality assessment was conducted by the Hayden’s criteria. Results A total of twenty one trials including 5229 patients were identified for the meta-analysis. 343 patients displayed BRAF mutations of 4616 (7.4%) patients with known BRAF status. Patients with BRAF wild-type (WT) showed decreased risks of progression and death with an improved PFS(HR 0.38, 95% confidence intervals 0.29–0.51) and an improved OS (HR 0.35 [0.29–0.42]), compared to BRAF mutant. In KRAS WT population, there were even larger PFS benefit (HR 0.29[0.19,0.43]) and larger OS benefit (HR 0.26 [0.20,0.35]) in BRAF WT. A response benefit for BRAF WT was observed (RR 0.31[0.18,0.53]) in KRAS WT patients, but not observed in unselected patients (RR 0.76 [0.43–1.33]). The results were consistent in the subgroup analysis of different study types. Heterogeneity between trials decreased in the subgroup and explained by sensitivity analysis. No publication bias of ORR, PFS and OS were detected. Conclusions The results indicate that BRAF mutant is a predictive biomarker for poor prognosis in mCRC patients undergoing anti-EGFR MoAbs therapy, especially in KRAS WT patients. Additional large prospective trials are required to confirm the predictive role of BRAF status.


Introduction
Colorectal cancer is the third mostly common human malignant tumor and is one major cause of cancer mortality in the western world [1]. Metastatic tumors account for 40% to 50% of newly diagnosed patients [2]. The prognosis of metastatic colorectal cancer(mCRC) remains poor. The introduction of targeted Epidermal Growth Factor Receptor (EGFR) Monoclonal Antibodies (MoAbs), namely Cetuximab and Panitumumab, has distinctly improved Overall response rate (ORR), Progression free survival (PFS) and Overall survival (OS). EGFR is a transmembrane tyrosine kinase receptor,which mediates the processes of proliferation, angiogenesis and invasion of cancer cells [3]. However, only 10%-20% of patients with mCRC can achieve benefits from anti-EGFR MoAbs [4]. EGFR expression is reported to be not correlated with clinical efficacy [5]. The benefit of targeted agents may attribute to the inhibition of its downstream signaling pathways, mainly RAS-RAF-MAPK and P3IK-PTEN-AKT [6]. Increasing evidences show that KRAS mutations at codons 12 and 13 in mCRC are predictive biomarkers of resistance to anti-EGFR MoAbs [7]. But KRAS mutations account only for 35% to 45% of nonresponders [8].
Recently, BRAF mutation (.95% of BRAF point mutations occure at BRAF V600E [9]) is introduced to be associated with resistance to targeted agents [10]. BRAF protein, a serinethreonine kinase, is the principal downstream molecular of KRAS [11]. A meta-analysis by Bokemeyer C, et al, in 2012 [12] based on two RCTs (the OPUS and CRYSTAL trials) reported that in KRAS wild-type(WT) patients, adding cetuximab to chemotherapy was beneficial for BRAF WT patients, but not for BRAF mutant patients. Another systematic review by Mao C, et al, in 2011 [13] found a response benefit for BRAF WT in KRAS WT patients, but found no response benefit for BRAF WT in unselected patients. And there is no meta-analysis for direct comparisons of PFS and OS between BRAF mutant and BRAF WT in mCRC patients using anti-EGFR MoAbs.
Here we aimed to provide a comprehensive, unbiased pooled analysis including ORR (risk ratio [RR] in patients with mutant BRAF versus(vs) these with WT BRAF) for response, PFS and OS (hazard ratios [HR] in patients with WT BRAF vs mutant BRAF) for progression and survival in patients with mCRC receiving anti-EGFR MoAbs therapies.

Search Strategy
We searched Pubmed, Web of Knowledge, the Cochrane library, and OVID without language limitation. The last search update was January 31, 2013. The search strategy mainly included three parts: (1) terms suggestive of ''BRAF'': (ie, ''BRAF'' or ''RAF''). (2) . Article types were restricted to clinical trials or Randomized Controlled Trials (RCT) in human. To ensure all related studies enrolled, we hand-searched several years of major journals such as ASCO (American Society of Clinical Oncology), ASCRS (American Society of Colon and Rectal Surgeons) and JCO (Journal of Clinical Oncology). The reference lists of primary studies and previous meta-analysis were scrutinized for additional publications.

Inclusion and Exclusion Criteria
The potential trials were screened for the following criteria: (1) patients with mCRC treated with cetuximab or panitumumab based therapy; (2) evalutaing BRAF mutations in the majority of patients and the number of patients with mutated BRAF was no less than one; (3) reported one or more indicators (including ORR, PFS and OS) to compare the prognosis of patients with WT BRAF to these with mutant BRAF; (4) retrospective trials, prospective trials, or randomized controlled trials. Trials evaluating progression with time to tumor progression (TTP), when TTP was defined as the time from the initiation date of Cetuximab or Panitumumab containing therapy to the first radiographic evidence of disease progression or death, were also included. We exlcuded trials without complete data, trials still in progression, and these without full text articles online. When reports overlapped or repeated, we retrived the data with longest follow up.

Data Extraction and Definitions
Data were extracted including the first author, publication year, patient baseline characteristics, the number of patients analyzed in the study, the number of patients with known BRAF mutation, the percentage of patients with Eastern Cooperative Oncology Group (ECOG) performance status #1point, the proportion of liver only metastasis, study design, line of treatment, chemotherapy regimens, anti-EGFR MoAbs used and the response criteria. For clinical outcome, we collected the number of responders for calculating RRs and 95% estimation intevals for ORR. We also extracted HRs and 95% credibility intervals for PFS and OS. If separate HR was not provided, we estimated HR and its variance from published survival curves by previously described methods and models [14,15]. Adjusted HRs and estimation intervals were also collected when reported.
Overall response included complete response (CR) and partial response (PR), non-response consisted of stable disease (SD) and progression disease (PD) according to the Response Evaluation Criteria in Solid Tumors (RECIST) [16] or World Health Organization (WHO) criteria [17]. PFS was defined as the time from the initiation date of anti-EGFR moAbs therapy to first evidence of disease progression or death of any cause, OS was defined as the time from the initiation date of anti-EGFR moAbs therapy to death of any cause. Outcome data were extracted separately in unselected population and in KRAS WT population. All data above were extracted by two independent investigators. When discrepancies existed, discussions were made to reach a consensus.

Assessment of Study Quality
For assessing the risk of bias in individual study, we used the Hayden's criteria to assess the quality [18]. This is based on six domains of potential study biases which should be included in a review of prognostic studies: study participation, study attrition, measurement of prognostic factors, measurement of confounding variables, mesurement of outcomes, and analysis methods. The criteria is not scored, but we designed a scoring scale based on the Hayden's criteria with some modifications to this study to quantize the assessment. The maximum score for each item was 2. Studies scoring 10-12 were defined as high quality, while these scoring 0-9 were considered low quality, just as previously defined by Maan ZN, et al. [19] (Table S1).

Statistical Analysis
We described statistics for baseline characteristics across eligible studies. A Risk ratio (RR) of ORR was calculated by the formula ½a=(azb)=½c=(czd) (a, b represented for the numbers of responders and nonresponders in BRAF mutant; c,d represented for the numbers of responders and nonresponders in BRAF WT in the same arm ) [20]. A HR and its variance were used directly if the trials provided. If not appropriate for direct analysis, we converted a HR and variance according to previous reported methods [14,15]. When not reported, a HR was estimated indirectly from other statistics such as log rank p value or calculated from published Kaplan-Meier survival curves by methods and models previously mentioned [14,15,21]. A RR,1 for response (BRAF mutation vs BRAF WT), and HRs,1 for PFS and OS (BRAF WT vs BRAF mutation) revealed poorer pognosis of patients with mutated BRAF over these with WT BRAF in anti-EGFR treatments.
Between-trial heterogeneity was assessed by both Q 2 statistic and I 2 statistic for more reliability. For Q 2 statistic, significant heterogeneity existed when p value was less than 0.10 [22]. For I 2 statistic, values above 50% were deemed to suggest large heterogeneity; values between 25%-50% indicated modest heterogeneity; values below 25% meant low heterogeneity [23]. But the values could be largely uncertain when few trials were pooled.
If the results of Q 2 statistic and I 2 statistic were conflicted, the conclusion of I 2 statistic was adpoted. The effect sizes across trials, namely pooled HRs and RRs, were estimated using the fixedeffect model by Mantel-Haenszel method when no significant heterogeneity existed(X 2 test: p$0.10). A random-effect model by Dersimonian and Laird method was adopted when there was a noted heterogeneity (X 2 test: p,0.10) [22,24]. The source of heterogeneity was explored by sensitivity analysis when large heterogeneity was presented. All p values reported were two-sided.
Publication bias were assessed by Egger's test (P,0.05 represented existing publication bias) and were reflected by visual symmetry of Begg's funnel plot on the natural logarithm of RRs or HRs [25].
Prespecified subgroup analysis was conducted in KRAS WT patients, as increasing evidence suggested KRAS mutation to be a predictor for resistance to anti-EGFR MoAbs therapy [26]. Subgroup analysis was also conducted according to different study types such as retrospective, prospective trials and RCT in both unselected population and KRAS population, to explore whether the results of meta-analysis from different study types being consistent. Sensitivity analyses were performed to evaluate the stability of pooled results by deleting one trial each time. The source of heterogeneity was also explored when strong heterogeneity between-trial existed. All the statistical analyses in the metaanalysis were performed with STATA software, version 11.0 (Stata Corporation, College Station, TX, USA, http://www.stata.com).

Study Selection and Characteristics
Total 318 potentially relevant records for retrieval were identified from Pubmed (n = 55), Web of Knowledge (n = 32), the Cochrane Library (n = 14), and OVID (n = 217). After reading headings and abstracts, 251 records were excluded. The remaining 67 full-texts articles were assessed for eligibility. We excluded 51 studies which did not meet eligibility criteria. Five additional trials were identified by manually searching the preference lists of previous meta-analysis, major meetings, primary studies and major journals. Finally, 21 eligible trials were included into the meta-analysis ( Figure 1).
In the subgroup analysis of different study types in unselected population, we performed meta-analysis separately according to retrospective, prospective trials and RCT. And the results were mostly consistent with the overall findings.There were still benefits for BRAF WT patients on PFS in no matter retrospective trials (HR 0.35, [0.24-0.52], p,0.001, Figure 2B Figure 2A) with significant heterogeneity (p = 0.022, I 2 = 81.1%, Figure 2A) (only two RCTs were included into the meta-analysis, sensitivity analysis to explore the the source of heterogeneity between studies was listed in Table S2).
We then performed meta-analysis of ORR, PFS and OS seperately in KRAS WT patients. There were a PFS benefit in BRAF WT (HR 0.29, [0.19-0.43], p,0.001), although there were considerable differences between the trials (Heterogeneity p = 0.033, I 2 = 56.2%, random effect model, Figure 3B). In the subgroup analysis of retrospective trials, heterogeneity decreased to still above 50% (p = 0.059, I 2 = 53.0%, Figure 3B). In sensitivity analysis, we tried to explore the source of heterogeneity from study quality, age and sex, but we didn't find out the source. Heterogeneity across trials may come from others (Table S2). We also conducted sensitivity analysis by deleting one study each time and the results were still consistent (results not provided in the study), which revealed the stability of the conclusion.There was also evidence of an OS benefit in BRAF WT patients (HR 0.  Figure 3C) without considerable differences across trials (Heterogeneity p = 0.764, I 2 = 0.0%, Figure 3C) and RCT (HR 0.31, [0.17-0.56], p,0.001, Figure 3C) were observed. There were also ORR benefits for BRAF WT in retrospective trials (RR 0.20, [0.08-0.52], p = 0.001, Figure 3A) with no significant variation between trials (Heterogeneity p = 0.988, I 2 = 0.0%, Figure 3A), and a benefit in RCTs (RR 0.38 [0.20-0.73], p = 0.004, Figure 3A).

Discussion
We performed the meta-analysis for the prognostic effects of anti-EGFR moAbs on mCRC patients with WT or mutant BRAF, which were based on the results of 21 eligible trials. The overall rate of BRAF mutation (7.4%) was similar to previously reported series [48]. The results demonstrated that patients with BRAF WT had decreased risks of progression (PFS: HR 0.38, p,0.001) and death (OS: HR 0.35, p,0.001) than patients with BRAF mutant. However, evidence of increased response in patients with BRAF WT was not enough (p = 0.328) comparing to BRAF mutated patients. In the subgroup analysis of different study types, there were still benefits for PFS and OS, but also not enough evidence of a response benefit for BRAF WT patients. In KRAS WT patients, results showed patients with BRAF WT not only decreased the risks of progression (PFS: HR 0.29, p,0.001) and death (OS: HR 0.26, p,0.001), but also increased responses (p,0.001) over these with BRAF mutant. Results were still consisted in the subgoup analysis of differents study types. Differences decreased in subgroup analysis and the conclusions didn't change in the sensitivity analysis.
Although a previous meta-analysis from Mao C, et al [13] demonstrates larger response benefit of anti-EGFR MoAbs in BRAF WT patients over BRAF mutant patients, they did not compare other indicators such as PFS and OS. Another published meta-analysis based on the OPUS and CRYSTAL trials from Bokemeyer C, et al, [12] indicates that adding Cetuximab to chemotherapy in mCRC is beneficial for BRAF WT patients, not for BRAF mutant patients. However, the study involves only two RCTs and is conducted only in KRAS WT patients. Direct comparison between BRAF mutant and BRAF WT on effects of MoAbs is also not reported. In this meta-analysis, we made direct comparison on effects of MoAbs between patients with mutant and WT BRAF. Generally speaking, our results confirmed that mCRC patients with BRAF mutant treated with MoAbs have poorer prognosis than these with BRAF WT, especially in KRAS WT population.
We know the limitations of our meta-analysis. Firstly, retrospective trials were also included, which may cause selective bias. Secondly, only four trials reported HRs and variances as we wanted. We had to calculate or convert HRs and variances for other trials from reported survival curves, which may introduce unavoidable bias. Thirdly, size effects from retrospective and prospectived trials are unadjusted, whilst size effects from RCTs are adjusted by patient baseline characteristics. Because individual patient data was not available, we conducted meta-analysis based on unadjusted and adjusted estimations,which may introduce confounding bias. Finally, the first and second end points were incosistant across different trials, so we didn't define them in this review.
Despite these limitations above, we confirm the conclusion that BRAF mutant is a predictive biomarker for poor prognosis in mCRC patients receiving anti-EGFR MoAbs therapy, especially in patients with KRAS WT. Therefore, screening for BRAF WT may promote the selection of potential mCRC patients whom will benefit from anti-EGFR moAbs.     Author Contributions