Prognostic value of EGFR and KRAS in circulating tumor DNA in patients with advanced non-small cell lung cancer: a systematic review and meta-analysis

EGFR (exon 19 and exon 21) mutations in patients with advanced non-small cell lung cancer (NSCLC) treated by EGFR-TKIs are associated with a better survival; while KRAS mutations predict a worse prognosis. However, there are divergent findings regarding the prognostic value of EGFR and KRAS mutations in circulating tumor DNA (ctDNA). We aimed to summarize the evidence for the use of circulating EGFR and KRAS mutations as prognostic factors in advanced NSCLC patients. We searched the network databases for studies reporting progression-free survival (PFS) and overall survival (OS) stratified by EGFR or KRAS mutations in ctDNA in advanced NSCLC patients. Thirteen studies enrolling 2,293 patients were reviewed. Correlation of circulating EGFR or KRAS mutations with patients’ prognosis was assessed by meta-analysis. The pooled analyses showed that EGFR mutations in ctDNA significantly prolong PFS (HR=0.64,95% CI 0.51-0.81, I2=0%, p=0.0002), namely, in patients treated by EGFR-TKIs. There is a trend to have a prolonged OS for advanced NSCLC patients with circulating EGFR mutations who were treated by EGFR-TKIs (HR=0.79, 95% CI 0.52-1.21, I2=0, p=0.28). KRAS mutations detected in ctDNA predict a worse PFS (HR=1.83, 95% CI 1.40-2.40, p<0.0001) and OS (HR=2.07, 95% CI 1.54-2.78, p<0.00001) in advanced NSCLC patients treated by chemotherapy. Sensitivity analyses and subgroup analyses demonstrated the stability of our conclusion. Our analysis showed that EGFR mutations in ctDNA predicted a better PFS, in particular in advanced NSCLC patients treated by EGFR-TKIs. KRAS mutations in ctDNA indicated a worse PFS and OS in patients treated by chemotherapy.


INTRODUCTION
Non-small cell lung cancer (NSCLC) remains the major cause of cancer-related mortality. Studies showed that epidermal growth factor receptor (EGFR)-tyrosine kinase inhibitors (TKIs) confer better outcome in patients with EGFR mutations (exon 19 deletions, exon 21 L858R point mutations) than in those with the wild type [1]. About 5-15% of NSCLC patients harbor EGFR mutations [2]. KRAS mutations predict worse prognosis among NSCLC patients treated by EGFR-TKIs or chemotherapy [3,4]. KRAS mutations are detected in about 30% of NSCLC in white people [5]. Approximately 97% of KRAS mutations in NSCLC involve codon 12 or codon 13 [3,6]. Several studies performed systematic review and metaanalysis to assess the prognostic value of EGFR and KRAS mutations in tumor tissue in NSCLC patients [4,[7][8][9].
Circulating tumor DNA (CtDNA) is shed into the bloodstream by tumor cells [10]. Evidence shows that ctDNA might be used as a noninvasive blood biomarker in tumor medicine [11,12]. Diagnostic tests for ctDNA such as OncoBEAM ® RAS CRC Kit (Sysmex Inostics GmbH), cobas ® EGFR Mutation Test V2 (Roche) and EGFR Mutations Detection Kit (AmoyDx) are commercially Review available for ctDNA detection. The published papers offered divergent findings regarding the prognostic value of EGFR and KRAS mutations in ctDNA in patients with advanced NSCLC. Only one study by Mao et al. conducted meta-analysis to explore the prognostic value of EGFR in ctDNA in advanced NSCLC patients [13]. However, overlapping studies were included in their study [13]. No studies did a systematic review and meta-analysis to assess the prognostic value of KRAS mutations in ctDNA in patients with advanced NSCLC. Thus, we conducted a systematic review and meta-analysis to explore their prognostic values in advanced NSCLC patients.

Included studies
A total of 2,295 potential studies were identified. After screening by title and abstract, 2,216 studies were excluded. The main reasons for exclusion were duplicative studies, reviews, not human studies, not relevant to ctDNA, incorrect tumor type and epigenetic alterations. Of the remaining 79 studies, the full text was screened and 66 studies were excluded for lack of follow-up, no information about prognosis, not restricted to advanced NSCLC patients, non-English literature, not restricted to KRAS or EGFR (exon19 and exon21) mutations. Finally, 13 studies met the inclusion criteria and were included for systematic review and meta-analysis (Figure1).

Study characteristics
Thirteen studies containing 1,452 patients were published between 2005 and 2013. These studies analyzed the relationship between circulating EGFR and KRAS mutations status and survival outcomes. The mean number of patients for each study was 67, ranging from 22 to 308. Four studies were retrospective and 9 were prospective. All studies were published in full and all had enough information to obtain the hazard ratios (HRs) and associated 95% confidence intervals (CIs). The main characteristics of the included publications are summarized in Table 1.

QUALITY ASSESSMENT
We assessed risk of bias using the Cochrane Collaboration's tool (the Cochrane Collaboration's tool for assessing risk of bias in randomized trials) [14]. All the included studies had a low risk of bias, as summarized in Figure 2.

Sensitive analysis
Sensitivity analysis by "leave-one-out" strategy showed that there was no "dominant" study driving the results of meta-analysis ( Figure 3B).

Subgroup analysis
We performed subgroup analyses on the basis of detection matrix (serum vs. plasma) and study year (prior to 2010 vs. after 2010). There was no statistical significance between these subgroups ( Table 2).
Subgroup analysis Subgroup analyses on the basis of detection matrix (serum vs. plasma) and study year (prior to 2010 vs. after 2010) were performed. There was no statistical significance between these subgroups (Table 3).

Sensitive analysis
Three out of 4 studies explored KRAS mutations in ctDNA as a predictive factor for the effect of chemotherapy on advanced NSCLC patients [3,6,24]. The pooled analysis showed that KRAS mutations in ctDNA indicate a worse PFS in patients treated by chemotherapy (HR = 1.81, 95% CI 1.38-2.38, I 2 = 17%) ( Figure 5B).
The "leave-one-out" strategy showed that there was no dominant study driving the results of meta-analysis (Supplementary Figure S3) Subgroup analysis We performed subgroup analysis based on detection sample (serum vs. plasma). There was no statistical significance between these subgroups ( Figure 5C).
The "leave-one-out" sensitivity analysis showed that our results were stable (Supplementary Figure S4).

Subgroup analysis
We performed subgroup analysis based on detection sample (serum vs. plasma). The pooled analysis demonstrated that the association between OS and KRAS mutation status was slightly stronger in plasma (HR = 2.29, 95% CI 1.65-3.16) than in serum (HR = 1.29, 95% CI 0.64-2.62), though the subgroup difference was not significant (p = 0.15) ( Figure 6C).

DISCUSSION
EGFR genotyping has become a routine test for the selection of patients as candidates for TKIs therapy. KRAS mutations test may help to predict less benefit of treatment with EGFR-TKIs and chemotherapy.
EGFR and KRAS mutation tests are often carried out from formalin-fixed paraffin embedded (FFPE) tumor tissue samples. However, molecular testing with FFPE has some limits. DNA extracted from FFPE is fragmented and also contains DNA lesions (including uracil and thymine deriving from cytosine deamination) that may lead to sequence artifacts [27]. Frozen tumor biopsies and fresh biopsies may complement FFPE. However, sometimes this is not feasible in clinical practice for advanced NSCLC, especially for successive dynamic monitoring. Tumor heterogeneity is another hurdle when utilizing tissue samples for mutation tests [28]. CtDNA may overcome these limits, and may be used as a surrogate [29]. However, we still have a limited understanding of the origination of ctDNA. If the ctDNA analyzed originates from apoptotic/necrotic tumor cells, it may have limited applicability for prognostic analysis. If the ctDNA utilized is from actively shedding tumor cells (either in circulation or in primary/metastatic sites), it may provide valuable information for treatment decision and prognostic prediction. Due to tumor heterogeneity, the choice of target treatment should ideally be based on the ctDNA from metastatic sites instead of on that from the primary sites [30]. Therefore, the limited understanding of its origination arises the question whether EGFR or KRAS mutations detected in ctDNA could be used as biomarkers for prognosis prediction.
The current meta-analysis established that advanced NSCLC patients with ctDNA EGFR mutations predicted a better PFS, namely, in patients treated by EGFR-TKIs. There is a trend to have a prolonged OS for advanced NSCLC patients with ctDNA EGFR mutations who were treated by EGFR-TKIs. Mao et al. carried out metaanalysis in advanced NSCLC patients treated by TKIs and arrived at a different conclusion. Their analyses showed that EGFR mutations detected in blood were associated with better OS (HR = 0.71, 95% CI 0.50-0.99, p = 0.61) [13]. The difference was likely to be caused by including suspected overlapping studies in that study [13]. Moreover, most of the HR estimates were extrapolated from the survival curves, which also contributed to this discrepancy. Huang et al. and Lee et al. carried out metaanalyses in NSCLC patients with tissue EGFR mutations. They found that the improvement in OS was only marginally statistically significant in patients receiving TKIs therapy, which was similar to our conclusion [7,8].
Our meta-analysis showed that at least in patients treated by chemotherapy, circulating KRAS mutations correlated with worse PFS and OS. This finding was consistent with the meta-analysis conducted by Chen et al, who evaluated the prognostic value of tissue KRAS mutation status [4].
Either serum or plasma was used as detecting matrix for genotyping in the eligible studies. The amounts of cell-free DNA are much higher in serum due to cell lysis during sample processing, hence reducing the fraction of tumor DNA in serum. However, this discrepancy did not alter our conclusions.
This meta-analysis had some advantages. First, we performed a comprehensive review and reported the most up-to-date published data. Second, no heterogeneity was found in this meta-analysis. Finally, this was the first metaanalysis to assess the prognostic value of circulating KRAS mutations in advanced NSCLC patients.
Despite our efforts to provide an accurate and comprehensive analysis, the limitations of our metaanalysis should be highlighted. First, EGFR exon 19 and exon 21 respond differently to TKIs treatment [9]; thus, it is necessary to perform subgroup analysis according to EGFR mutation subtypes in a future study. Second, we did not perform subgroup analyses based on age, sex, smoking status, and detection methods due to insufficient data. Another limitation of our meta-analysis is that the status of other actionable mutations such as ALK rearrangements were not considered and should be included in an analysis with more available data in future studies. Finally, the eligible studies only performed univariate analyses; we cannot infer from our meta-analysis whether EGFR or KRAS mutations in ctDNA could be an independent factor or not.
Selection bias may exist in our paper. For EGFR and KRAS involved in lung cancers, striking differences in molecular alterations of these genes have been found in never and ever smokers [31]. Epidemiological studies of lung cancer showed that EGFR mutations occur more frequently in never smoker East Asia, while KRAS mutations occur more frequently in smokers and less common in never smoker East Asia [30]. In our paper, the rate of EGFR mutations is higher than usually expected, and the rate of KRAS mutations is much lower in the selected studies. This may be due to a high rate of neversmoker patients in our study. Despite the aforementioned limitations, this metaanalysis suggested that EGFR mutations detected in ctDNA were associated with a better PFS, namely, in patients treated by EGFR-TIKs. There is a trend to have a prolonged OS for patients with ctDNA EGFR mutations who were treated by EGFR-TKIs. Circulating KRAS mutation-positive status in advanced NSCLC predicts a worse PFS and OS in patients treated by chemotherapy.

Search methods for identification of studies
Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines were conformed to identify potential relevant studies. We did systematic electronic searches of Medline, Embase, Web of Science, the Cochrane Library, and Scopus up to October 10, 2015 (no start date limit was applied). The search strategy used was as follows: "Carcinoma, Non Small Cell Lung", "Carcinomas, Non-Small-Cell Lung", "Lung Carcinoma, Non-Small-Cell", "Lung Carcinomas, Non-Small-Cell", "Non-Small-Cell Lung Carcinomas", "Non small Cell Lung Cancer", "Non-Small-Cell Lung Carcinoma", "Non Small Cell Lung Carcinoma", "Carcinoma, Non-Small Cell Lung", "NSCLC", "Non-Small Cell Lung Cancer", "ctDNA" , "circulating tumor DNA" , "cell free DNA", "serum DNA", "plasma DNA" , "circulating DNA, free DNA", "free DNA", "cfDNA", "prognosis", "survival", "prognostic", "predictive". Relevant MeSH (Medline) or Emtree (Embase) terms were used where possible. We also hand searched the relevant reference lists to identify new studies. Conference posters and letters that fulfilled the inclusion criteria were also included to capture grey literature. The literature search was confined to English publications.
Two investigators (Fan GW and Zhang K) independently assessed each study for inclusion, and discrepancies were resolved by discussion. Whenever overlapping samples existed (e.g., same authors, overlapping period of study, same protocol ID, overlapping patients), we retained the report with the largest patient population.

Criteria for considering studies for this review
Eligible studies met the following criteria: (1) dealt with advanced NSCLC (stage IIIB or IV) patients only; (2) analyzed the correlation between patient survival and EGFR mutations (exon 19 deletions or L858R) and/or KRAS mutations in ctDNA; (3) had follow-up for overall survival (OS) and/or progression-free survival (PFS); and (4) provided enough information to obtain HRs directly or indirectly.
Both prospective and retrospective studies were included. Reviews, comments, and case reports were excluded. Studies with less than five patients were also excluded.

Data extraction
Two investigators (Fan GW and Ding JS) independently screened the eligible studies and extracted data using a predefined information sheet that included the following information: first author, publication year, country where the study was conducted, publication type, study design, patients included in analysis, median age, percentage of males, percentage of non-smokers, tumor stage, treatment, detection sample, the time of sampling, ctDNA detection, detection methods, and number of patients with positive ctDNA. Discrepancies were resolved by discussion.

Quality assessment
We evaluated the quality of included studies using the Cochrane Collaboration's tool for assessing risk of bias [32]. Specifically, studies were judged on (1) selection bias: studies that had an explicit statement of inclusion and exclusion criteria were rated as low risk of selection bias; (2) accuracy of exposure assessment, also called measure bias: studies with an explicit statement regarding ctDNA detection methods were rated as low risk; (3) accuracy of other variable assessment: other molecular alterations such as epigenetic alterations and EGFR T790M also influence the prognosis of NSCLC patients; (4) accuracy of outcome assessment; (5) bias due to incomplete follow-up (e.g., median follow-up length, range, and loss-to-follow-up rate were satisfactorily reported); (6) confounding, which included known or commonly discussed confounders in the relationship between ctDNA and survival, such as age, smoking, or other factors that were adjusted.

Measures of treatment effect
The primary outcome was PFS. The secondary outcome was OS. Two investigators (Fan GW, Zhang K) extracted the HRs and their 95% confidence interval (95% CI) to assess the prognostic value of EGFR and KRAS mutations in ctDNA. If HRs for ctDNA were not available, we calculated them indirectly using the methods of Parmar [33]. By convention, HR = 1 indicates a lack of association between ctDNA status and prognosis; HR > 1 indicates a worse survival for patients with ctDNA positive; HR < 1 represents a benefit outcome for the ctDNA-positive group.

Statistical analysis
We used the chi-square test to detect heterogeneity and I 2 statistic to measure heterogeneity. A p value > 0.10 and an I 2 < 50% indicated a lack of significant heterogeneity; then, the fixed-effects model was used to calculate the pooled HR. Otherwise, the random-effects model was adopted. The effect of ctDNA status on survival was considered statistically significant if the 95% CI for the overall HR estimate did not contain 1.
Sensitivity analyses were conducted for the metaanalysis to check for stability of the overall results. Subgroup analyses were performed using a random-effects model because of the diverse clinical heterogeneity.
Analysis was carried out using Review Manager 5.3 and Stata 12.

Ethics statement
This study was a literature-based study and no ethics approval was needed.