Diagnostic and prognostic value of blood samples for KRAS mutation identification in lung cancer: a meta-analysis

Circulating tumor DNA (ctDNA) and tumor cells (CTC) are novel approaches for identifying genomic alterations. Thus, we designed a meta-analysis to evaluate the diagnostic value and prognostic significance of a KRAS proto-oncogene, GTPase (KRAS) mutation for lung cancer patients. All included articles were from PubMed, EMBASE, Web of Science and Cochrane Library. Twelve articles that described 1,131 patients were reviewed. True positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) were used to calculate pooled sensitivity, specificity, the positive likelihood ratio (PLR), the negative likelihood ratio (NLR), a diagnostic odds ratio (DOR), the area under the curve (AUC) and corresponding 95% confidence intervals (95% CI). PLR is calculated as sensitivity/(1-specificity) and NLR is (1– sensitivity)/specificity. DOR is a measured of diagnostic effectiveness (PLR/NLR). A survival analysis subgroup was also designed to evaluate prognostic significance. Pooled sensitivity, specificity, PLR, NLR, DOR and AUC were 0.79 (95% CI, 0.63-0.89), 0.93 (95% CI, 0.89-0.96), 12.13 (92% CI, 7.11-20.67), 0.22 (95% CI, 0.12-0.41), 54.82 (95% CI, 23.11-130.09), and 0.95 (95% CI, 0.93–0.96), respectively. KRAS mutation and wild-type hazard ratios for overall survival and progression-free survival were 1.37 (95% CI, 1.08–1.66), 1.46 (95% CI, 1.15-1.77) in blood samples, and 1.16 (95% CI, 1.03–1.28), 1.28 (95% CI, 1.09–1.46) in tumor tissue.


INTRODUCTION
Cancer is a serious global public health problem and lung cancer, in particular, is a leading cause of cancerrelated death in the United States. In 2016, almost 250,000 new cancer cases will be reported and slightly more than 150,000 deaths will result. [1] Additionally, lung cancer is the chief cause of cancer death among men and the second most common cause of cancer death among women worldwide. [2] Such high mortality is due to lack of early detection using lung cancer markers.
KRAS, one of the most frequently mutated oncogenes, contributes to the mitogen-activated protein (MAP) kinase pathway, which controls cell growth and differentiation. [3,4] The KRAS pathway is also involved in the regulation of lung cancer, participating in the downstream signaling network of epidermal growth factor receptor (EGFR). The most commonly mutated codons are 12, 13, and 61 and this causes drug resistance to EGFR tyrosine kinase inhibitors (EGFR-TKIs). Several studies suggest that KRAS mutations should be known prior to using EGFR-TKI therapy for lung cancer patients. [5][6][7] Although tumor tissue is the reference standard for KRAS mutation confirmation, obtaining tissue samples is difficult, costly, and invasive. [8] In addition, most advanced lung cancer patients are unable to tolerate Review Oncotarget 36813 www.impactjournals.com/oncotarget surgical procedures. Thus, a more feasible but accurate method for assaying KRAS mutations is needed. Blood testing is less invasive, easily-accessible and can be repeated. [7,9] Thus, ctDNA and CTCs can be used as a high diagnostic value and prognostically significant source for identifying KRAS mutations in lung cancer patients.

Baseline characteristics of identified studies
Baseline characteristics of eligible studies are shown in Table 1. The included articles were published between 2003 and Jan, 2017. Two articles had more than one combination of statistics. [13,15] CTC were   Figure S1, and sensitivity analysis is presented in (Figure 2A) which was accomplished by excluding studies one by one. Data were stable and were not significantly different.     line represents the 95% CI of each study, square proportional means the weight of every study. The weight is evaluated by the sample size and is presented as percent of total. The diamond represents pooled sensitivity, specificity and 95% CI.

Sub-groups
Sub-group analysis is shown in Table 2. Race, detection method and treatment are displayed and data show that Asian subjects experienced greater diagnostic accuracy compared with Caucasians. CTC and frozen tissue was more sensitive than ctDNA and FFPE.

Outcomes
The estimated pooled HRs for OS and PFS is displayed in Figure 6 and data show that poorer prognosis is correlated with KRAS mutations. Subgroup analysis indicated that lung cancer patients with KRAS mutations had a significantly shorter OS and PFS compared to wildtype lung cancer patients. Additionally, there was no significant difference between HRs for blood samples and tumor tissues so both can be used.  KRAS mutation identification. C. ROC plane for threshold effect. Each black spot represents an included study and does not constitute a "shoulder shape" graph, which represents no significant threshold effect.

Heterogeneity and publication bias
Oncotarget 36818 www.impactjournals.com/oncotarget specificity were 86.19 (95%CI, 73.88-98.49) and 64.71 (95%CI, 26.55-100.00), which implies a statistically significant heterogeneity. Most heterogeneity was derived from the threshold effect and differenced among studies. The ROC plane and statistical data show no significant threshold effect ( Figure 4C). The Spearman correlation coefficient was 0.367 and the P value was 0.197 (P > 0.05), indicating no significant threshold effect. Therefore, we suspect that heterogeneity is likely rooted in differences among studies. Potential publication bias was evaluated using a Deek regression test ( Figure 4A), and no significant publication bias was discovered (p = 0.218 > 0.05).

DISCUSSION
Detecting KRAS mutations in lung cancer is useful for predicting patient outcomes and targeting therapy and tumor tissue is currently used for this assay. Limitations to this approach include patient age and health, so a simple, minimally invasive approach for measuring KRAS mutations is required and blood sampling may be that solution. To address this issue, we conducted a metaanalysis to evaluate the diagnostic accuracy and prognostic significance of using blood samples for KRAS mutation assay. The results show that blood sampling offered high sensitivity and specificity which suggests that KRAS mutations can be assayed this way when tumor tissue is inconvenient or unavailable. Also, blood samples offered high diagnostic accuracy. [41,42] Finally, likelihood ratios and post-test probability are also important testing standards. [43] The value of likelihood ratios ranges from 0 to infinity. When likelihood is 2-5, post-test probability is slightly increased. When likelihood is > 10, post-test probability increases significantly. In this study, PLR was 12 and NLR was 0.22, which clearly changed the post-test probability.
Subgroup analysis to identify factors that can influence diagnostic accuracy included race, detection method, and tissue treatment. Data show that compared with Caucasians, KRAS mutations in blood samples of Asians was more accurate and sensitive when using Weight is evaluated by sample size and presented as percent of total. Diamond represents pooled sensitivity, specificity and 95% CI.
Oncotarget 36819 www.impactjournals.com/oncotarget and square proportional means the weight of every study. Weight is evaluated by sample size and is presented as percent of total. Diamond represents pooled HR and 95% CI.
Oncotarget 36820 www.impactjournals.com/oncotarget frozen tumor tissue samples and CTC methods compared to FFPE tissue samples and ctDNA. FFPE can lead to a cross-link between proteins and nucleic acids but this did not occurs with nitrogen-frozen tissues. CTC was more sensitive than ctDNA [44], perhaps due to fewer included studies. Detection methods, collection timing, and TNM stage were not analyzed due to too few studies including this information. Subgroup survival analysis indicated that KRAS mutations are associated with significant increases in mortality but there were no differences between blood samples and tumor tissues for OS and PFS, which suggest that blood sampling is suitable for replacing tissue assay.
This is the first meta-analysis to evaluate KRAS mutations in blood samples for treating lung cancer. Liquid biopsies allow identification of molecular targets, assessment of prognosis, monitoring therapeutic response and molecular profiles in real time as well as diagnosis of disease recurrence or progression. We found that liquid was highly accurate and high ctDNA and CTC are correlated with poorer prognosis for lung cancer patients. [45] Thus, ctDNA and CTC can be used to confirm KRAS mutations in lung cancer instead of tumor tissue and suggest details about prognosis. The diagnostic value and prognostic significance of blood sampling for lung cancer patient monitoring is unclear but our data suggest that it is worth investigating.
The meta-analysis has several limitations such as potential publication bias. We used well-selected articles and Supplementary Data and Deek's funnel plot did not confirm statistical significance (p = 0.170 > 0.05). Second, some studies were small and this may have caused bias but a sensitivity analysis suggested that sample size did not influence pooled results significantly. Third, significant heterogeneity existed in our meta-analysis and the ROC plane and Spearman correlation coefficient data indicated that heterogeneity was not due to a threshold effect. Thus, heterogeneity may be primarily due to small sample studies [12,17] and differences among study detection methods. Studies also differed with respect to race, TNM classification, and percent of lung adenocarcinomas. We tried to establish a subgroup for test methods but because we had few studies and varied methods within them, this was difficult. Future studies should be designed to evaluate differences in detection methods. Finally, in the prognostic analysis sub-group, most studies did not provide a HR so we calculated one (at 95% CI using a survival curve) and it may indicate result bias.
In conclusion, lung cancer is a leading cause of cancer-specific mortality around the world and with the rapid development of liquid biopsy, CTCs and ctDNA provide a novel method for assaying KRAS mutations in lung cancer. Our meta-analysis indicates that this approach has advantages over other methods and that it is highly specific, non-invasive, and a repeatable measuring approach with diagnostic and prognostic value that allows real-time monitoring.

Data source and search strategy
We reviewed reports published in PubMed, EMBASE, Web of Science and the Cochrane Library. We used these searched terms: 'KRAS' or 'GTPase KRAS' or 'V-Ki-ras2 Kirsten rat sarcoma viral oncogene homolog,' 'serum' or 'plasma' or 'circulating,' 'mutation,' 'cancer' or 'carcinoma' or 'tumor' or 'neoplasm,' and 'lung.' Only studies published in English were included.

Inclusion and exclusion criteria
Inclusion criteria for primary studies were d a definite diagnosis of lung cancer; KRAS mutations diagnosed with ctDNA, CTC, or tumor tissue; sufficient information for a 2 x 2 table. Articles were excluded if the KRAS mutations were not detected using tumor tissue; tumor tissues and blood samples were not matched; there was insufficient information reported to finish a 2 x 2 table; or lung cancer data were not separate from other cancer data. All selected studies were managed using EndNote X7. Studies included in our meta-analysis were assessed by two investigators independently.

Data extraction and quality assessment
The first author's name, year of publication, country, number of patients, sex ratio, the proportion of smokers included, adenocarcinoma (AC) ratio, tumor tissue treatment, use of serum or plasma, KRAS mutation detection methods, and TNM stage were collected from eligible studies. Then, 2 x 2 tables were designed to show TP, TN, FP, and FN. When a KRAS mutation was detected by multiple methods, data for all methods were extracted, recorded, and evaluated by two investigators independently. QUADAS-2 (quality assessment of diagnostic accuracy studies 2) was used to evaluate diagnostic accuracy quality [46] using patient selection, index test, reference standard, and flow and timing.
OS was defined as the survival time from randomization and PFS was defined as the time from randomization to progression, recurrence, death or termination of follow-up. When studies did not report HRs directly, two independent investigators calculated survival data from survival curves using an Engauge Digitizer, version 4.1integrated to calculate overall HR. The threshold effect was measured by using the ROC plane, a Spearman correlation coefficient and a p-value. We evaluated race, detection methods, and tissue treatment. Publication bias was measured using a Deek's funnel plot and (p = 0.218) which indicated no significant bias. [49] All statistical analyses were performed using STATA software (version 12.0, STATA Corp, MIDAS module) and Meta-Disc. Quality assessment was managed with Review Manager 5.3.