Detection of KRAS mutation using plasma samples in non-small-cell lung cancer: a systematic review and meta-analysis

Background The aim of this study was to investigate the diagnostic accuracy of KRAS mutation detection using plasma sample of patients with non-small cell lung cancer (NSCLC). Methods Databases of Pubmed, Embase, Cochrane Library, and Web of Science were searched for studies detecting KRAS mutation in paired tissue and plasma samples of patients with NSCLC. Data were extracted from each eligible study and analyzed using MetaDiSc and STATA. Results After database searching and screening of the studies with pre-defined criteria, 43 eligible studies were identified and relevant data were extracted. After pooling the accuracy data from 3341 patients, the pooled sensitivity, specificity and diagnostic odds ratio were 71%, 94%, and 59.28, respectively. Area under curve of summary receiver operating characteristic curve was 0.8883. Subgroup analysis revealed that next-generation sequencing outperformed PCR-based techniques in detecting KRAS mutation using plasma sample of patients with NSCLC, with sensitivity, specificity, and diagnostic odds ratio of 73%, 94%, and 82.60, respectively. Conclusion Compared to paired tumor tissue sample, plasma sample showed overall good performance in detecting KRAS mutation in patients with NSCLC, which could serve as good surrogate when tissue samples are not available.


Introduction
Lung cancer is a leading cause of cancer-related death worldwide (1). As its most prevalent subtype, non-small cell lung cancer (NSCLC) represents approximately 85% of lung cancer cases (2). Treatments of NSCLC include surgery, radiotherapy, chemotherapy, immunotherapy, and targeted therapy in tumors harboring certain oncogenetic variations, e.g., anti-epidermal growth factor receptor (EGFR) therapy (2).
Kirsten rat sarcoma viral oncogene homologue (KRAS) is the most frequently mutated oncogene in many types of cancer (3), with an overall prevalence of 27.5% in NSCLC (4). Mutation of KRAS gene is associated with resistance to anti-EGFR therapies (5)(6)(7). In addition, although KRAS was thought to be an "undruggable" target, it has become "druggable" after the successful approval of KRAS (G12C) inhibitor (Sotorasib) for the treatment of KRAS G12C-mutated metastatic NSCLC (8). Due to these important roles of KRAS mutation in targeted therapies, accurate detection of KRAS gene mutations, especially G12C, is crucial for the success of anti-EGFR therapies and KRAS inhibitors.
The detection of KRAS mutations in tumors is usually performed using tumor tissue samples, e.g., formalin-fixed paraffin-embedded (FFPE) tumor tissue samples. However, tissue samples are sometimes not available, or may not reflect the real-time mutation status of tumor due to the existence of cancer evolution (9). Research efforts were therefore made to find possible surrogates for tumor tissue samples, which are mainly cell-free DNA (cfDNA)-containing samples, such as plasma, urine, saliva, feces, exhaled breath condensate, and etc (10,11). Before their clinical application, however, those surrogate sample types needs to be validated for their accuracy performance in detecting KRAS mutations. Many such studies have been conducted. A recentlypublished systemic review and meta-analysis by Palmieri (12) summarized the results of 40 relevant studies and reported an overall adequate accuracy of cfDNA-containing samples. This metaanalysis by Palmieri focused on cfDNA, and involved studies using plasma, urine, or sputum samples. However, cfDNA levels in the three sample types are quite different, which could potentially influence accuracy performance. In addition, compared to urine or sputum samples which could be highly concentrated or diluted, cfDNA levels in plasma samples are considered to be more stable and therefore had potentially better stability in accuracy performance. Considering these advantages, we chose to focus on plasma, and aimed to better understand the accuracy performance of plasma sample in KRAS mutation detection in NSCLC, including potential impact of patient characteristics.

Literature searching and selection of publication
Literature search was performed by BY and JZ in June 2022. Online literature databases (Pubmed, Embase, Cochrane Library, and Web of Science) were searched using keywords: "KRAS", "plasma", and "NSCLC". Alternative spelling or abbreviations were also included in the literature search, e.g., non-small-cell lung cancer, non-small-cell lung carcinoma, NSCLCs, NSCLC's, plasmas, and plasma's (please see detailed searching strategy in Supplementary Material). Searching results were exported from each database. Duplicated literatures were then identified by matching titles, names of first author, or identification numbers (e.g., Pubmed ID) of literatures from different databases. After removing the duplicated literatures, the abstracts of the searching results were firstly screened to exclude irrelevant literatures. The full texts of the rest literatures were then downloaded and screened for eligible studies. The criteria used for the two screening steps were as follows. Inclusion criteria: all original studies testing KRAS mutation in paired plasma and tumor tissue samples of NSCLC. Exclusion criteria: 1) not a human study; 2) missing plasma or tumor tissue samples; 3) plasma and tumor tissue samples were not paired; 4) not testing KRAS mutation in either plasma or tissue samples; 5) lacking KRAS wild-type or KRAS mutated samples; 6) not an original study; 7) un-interpretable data; 8) not NSCLC samples. Accuracy data were then extracted from the KRAS mutation testing results of paired plasma and tumor tissue samples in the eligible studies, including numbers of true positive, false positive, false negative, and true negative. In addition, characteristics of patients or techniques were also extracted, including region and population of studies, tumor stage, and techniques used to test KRAS mutation in plasma and in tissue samples. All the eligible studies were evaluated by quality assessment of diagnostic accuracy studies 2 (QUADAS-2) (13). Any disagreement between the two investigators (BY and JZ) were solved by a third investigator (PC). PRISMA 2009 Checklist is included in Supplementary Material.

Statistical analysis
Statistical analysis was performed using Meta-DiSc 1.4 (14) and STATA 12.0 (STATA Corp.). Sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under curve (AUC) of summary receiver operating characteristic (SROC) curve were pooled from the accuracy data extracted from the eligible studies. During the pooling, random effects model was used when significant heterogeneity was observed (I 2 ≥ 50% and P < 0.05), and fixed effects model was used when no significant heterogeneity was observed (14). In case of significant heterogeneity, threshold analysis and meta-regression were performed to find its possible sources. Deek's funnel plot asymmetry test was performed to find potential publication bias in the eligible studies. P < 0.05 was considered statistically significant.

Search results
As shown in Figure 1

Review of eligible publications
Twenty-nine of the 43 eligible studies (Table 1) used nextgeneration sequencing (NGS) to detect KRAS mutation in plasma samples. In the rest 14 studies, 12 studies used PCRbased techniques, 1 study used pyrosequencing, and 1 study used MassARRAY.

NGS
In the eligible studies using NGS, sensitivities ranged from 25% to 100%, and specificities and concordance rates were relatively higher, ranging from 64% to 100% and from 52.63% to 100%, respectively.
The specificity and concordance rate were 99.24% and 99.32%, respectively. Similarly, study by Narayan (17) showed perfect matching (100% concordance rate) of KRAS mutation results between plasma and tissue samples. However, study by Paweletz (16) and by Couraud (18) showed much lower sensitivity (54.55% and 75%, respectively), although high specificity (100%) was observed. In the study by Wang Z (19), circulating singlemolecule amplification and resequencing technology (cSMART) showed sensitivity of 58.82%, specificity of 100%, and concordance rate of 93.20%. The large variations in the sensitivity of KRAS mutation detection in plasma samples may be due to the small number of patients included in these studies.

PCR-based techniques
A total of 4 studies used digital droplet PCR (ddPCR) to detect KRAS mutation in plasma samples (44)(45)(46)(47). Although ddPCR is a sensitive technique which could detect genetic mutations as low as 0.01%, the results of these studies did not show high accuracy of ddPCR in plasma-based KRAS mutation detection. Sensitivity ranged from 51.43% to 87.88%, and specificity ranged from 88.52% to 100%, resulting in concordance rates from 75% to 96%.
Other than ddPCR, several PCR-based techniques were also used to detect KRAS mutation in plasma samples, such as PANAmutyper, PCR-restriction fragment length polymorphism (PCR-RFLP), multiplex PCR, Amplification Refractory Mutation System (ARMS), and PCR/ligase detection reaction (LDR) technique. Overall, those PCR-based techniques were mostly used in early studies, which showed sensitivity ranging from 33.33% to 100%, specificity from 50% to 100%, and concordance rate from 55.56% to 100%.
PANAmutyper is a multiplex PCR method which increases sensitivity through suppressing amplification of wild-type DNA using specific peptide nucleic acids (PNA) (48). In the two studies using PANAmutyper, the sensitivity was 33.33% and 50%, and specificity was 100% and 89.43%, resulting in concordance rates of 88.89% and 85.93%, respectively (48, 49).
Multiplex PCR was used in two studies. Study by Zhang (52) used SurExam MEL (SurExam Biotech), a typical commercial multiplex PCR, to detect KRAS mutation in plasma samples, and sensitivity, specificity, and concordance rate were 33.33%, 98.80%, and 96.51%. In the study by Punnoose (53), the KRAS mutation results of plasma samples matched perfectly with tissue samples (100% concordance rate).
An early study by Mack (54) used KRAS Scorpion-ARMS test kit (DxS Ltd), and results showed 50% sensitivity, 100% specificity, and 97.96% concordance rate. Campos (55) and colleagues developed a microfluidic solidphase extraction device to extract cfDNA, which were then analyzed using PCR/LDR technique. Only 3 NSCLC samples were tested in the study, and the results showed 100% sensitivity, 50% specificity, and 66.67% concordance rate.
In all, the 43 eligible studies compared KRAS mutation status in paired plasma and tissue samples from 3341 NSCLC patients. Thirty-nine of the 43 eligible studies (39/43) showed high specificity (≥ 80%), and 37 studies showed high concordance rate (≥ 80%). However, high sensitivity (≥ 80%) was only observed in 14 out of 43 studies.

Quality assessment of eligible studies
Quality assessment of eligible studies was performed using QUADAS-2. As shown in Table 2, the 43 eligible studies showed overall good quality, with high risk observed in only 2 studies (both in flow and timing). In the assessment of risk of bias, percentage of low risk ranged from 46.51% (n = 20, Index test) to 69.77% (n = 30, both patient selection and reference standard). In the application concerns, no high risk was observed, and percentage of low risk ranged from 83.72% (n = 36, reference standard) to 86.05% (n = 37, both patient selection and index test).
Since significant heterogeneity (I 2 ≥ 50% and P < 0.05) was observed, we further analyzed its possible sources. Analysis of diagnostic threshold showed no significant threshold effect (spearman correlation coefficient = 0.058, P = 0.714). Metaregression revealed that inter-study heterogeneity was associated with techniques used for plasma sample (P = 0.0388), but not with techniques used for tissue sample (P = 0.1280), region of study (P = 0.3299), tumor stage (P = 0.3049), or race of patients (P = 0.7798).
Subgroup analysis was then performed on different techniques used for plasma sample. The 43 eligible studies were grouped into three subgroups: NGS, PCR-based techniques, and other techniques. Meta-analysis was performed in each subgroup except other techniques due to limited number (only two) of studies in that subgroup. As shown in Table 3 Twenty-four of the 43 eligible studies used late-stage (stage III and IV) NSCLC samples, and 13 studies used NSCLC samples of any stage (stage I to IV). As shown in Table 3, pooled accuracy results of the two subgroups (stage III-IV versus stage I-IV) did not differ much from each other. However, this result should be treated carefully because although early-stage NSCLC samples were involved, majority of the samples were still late-stage in stage I-IV subgroup. The rest 6 studies were not involved in the subgroup analysis, including 1 study using early-stage (I and II) NSCLS samples only, and 5 studies which did not disclose the tumor stage of samples. Majority of the 43 eligible studies were conducted using samples from Caucasian patients, and the rest studies used samples of Asian patients. Between the two subgroups, pooled accuracy data were similar (see Table 3).
Publication bias was evaluated using Deek's funnel plot (Figure 3). The results indicated no significant publication bias (P = 0.097).

Discussion
Before anti-EGFR therapies are given to NSCLC patients, it is important to determine whether the tumor carries KRAS mutation since it may lead to resistance to anti-EGFR therapies. Moreover, determination of KRAS mutation status is also required before the usage of KRAS (G12C) inhibitor, e.g., Sotorasib. Tumor tissue samples are the "gold standard" in the determination of KRAS mutation. However, tumor tissue samples are sometimes not available, and cfDNA-containing samples (e.g., plasma, urine, saliva, etc.) have been intensively investigated as surrogates for tissue samples. A recently-published systemic review and meta-analysis by Palmieri summarized the performance of cfDNA-containing samples in detecting KRAS mutation in NSCLC (12). Due to the higher and more stable levels of cfDNA in plasma compared to other cfDNAcontaining sample types, we focused solely on plasma in this systemic review and meta-analysis, and investigated its accuracy in determining tumor KRAS mutation status in NSCLC.
In order to investigate the accuracy of KRAS mutation detection using plasma samples, several previous studies compared KRAS mutation results in paired plasma and tissue samples from patients with NSCLC. After database searching and screening, we identified 43 eligible studies. After pooling the KRAS mutation status from 3341 patients with NSCLC, the results showed overall moderate sensitivity (0.71) and high specificity (0.94). Other important  Pooled sensitivity, specificity, DOR, and SROC curve of eligible studies.

FIGURE 3
Deek's funnel plot.  (12), the pooled sensitivity and specificity were 0.71 and 0.93, respectively, and DOR was 35.24, which were similar to the findings of our study. Since significant inter-study heterogeneity was observed during the pooling (I 2 ≥ 50% and P < 0.05), we investigated its possible sources. Analysis of diagnostic threshold did not indicate significant threshold effect. Meta-regression revealed significant association between inter-study heterogeneity and techniques used for plasma sample. This is different from Palmieri's study, in which detection method did not contribute to heterogeneity (12). No significant association was shown between heterogeneity and other covariates (techniques used for tissue sample, region of study, tumor stage, and race of patients).
Different from Palmieri's study, we further conducted subgroup analysis. Subgroup analysis on technique used for plasma sample was firstly performed. After pooling the accuracy results, we found that NGS outperformed PCR-based techniques in many accuracy parameters, including sensitivity (0.73), DOR (82.60), and AUC of SROC curve (0.9162). We further divided the group of PCR-based techniques into two groups: ddPCR and other PCR-based techniques. Compared to NGS, ddPCR showed similar sensitivity (0.68), specificity (0.97), and DOR (85.60), except for surprisingly low AUC of SROC curve (0.2741) which was possibly due to the limited number of studies in this subgroup (Table 3).
We also performed subgroup analysis on region of study. Studies performed in Asia showed the highest AUC of SROC curve (0.9381). Studies performed in America showed the highest sensitivity (0.76) and DOR (111. 35), and similar AUC of SROC curve with Asia (0.9272), indicating overall the highest accuracy of the studies from America.
Late-stage tumors was reported to be associated with significantly higher fraction of circulating tumor DNA (ctDNA) in cfDNA (58), which may indicate potentially better performance of genetic testing using these samples. In the 43 eligible studies, involvement of early-stage samples did not significantly influence the accuracy results. However, this result should be treated with care because numbers of early-stage samples were much smaller than late-stage samples in a large proportion of these studies. Race of patients also did not show significant impact on the accuracy results. The performance of KRAS mutation testing using plasma was similar between Asian and Caucasian patients. Significant publication bias was not observed using Deek's funnel plot asymmetry test.
In summary, results of this systemic review and meta-analysis indicated overall high accuracy of plasma samples in predicting KRAS mutation results of paired NSCLC tumor tissue samples. Plasma could serve as surrogates when tissue samples are not available, although it may miss a small proportion of patients carrying KRAS mutation considering its moderate sensitivity. Among different techniques, NGS showed the best accuracy. Although majority of accuracy results were comparable to NGS, ddPCR suffered from its low AUC of SROC curve. Therefore, NGS is recommended in the detection of KRAS mutations in plasma samples of patients with NSCLC, especially when multiple genetic variations are tested considering the high-throughput of the technology. Limitation of this study may be the small number of studies in the ddPCR subgroup and limited numbers of early-stage tumor samples used in some studies, which must be treated carefully. In addition, although different techniques are generally thought to have similar performance in tumor samples considering the high abundance of DNA, it may still cause potential bias. Large prospective studies are required to further validate the results of this study.

Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.