Liquid Biopsy as Surrogate to Tissue in Lung Cancer for Molecular Profiling: A Meta-Analysis.

BACKGROUND: The accurate microscopic diagnosis of lung cancer has become insufficient due to the concept of personalized medicine. Tissue samples are used not only for microscopic diagnosis but also for the assessment of the different targets. Biopsies are performed in 80% of the patients and they are not sufficient for molecular diagnosis in 30% of the cases. Liquid biopsy (LB) has been reported as a possible surrogate to tissue samples and has been introduced in the management scheme of the patients since 2014. We aimed to highlight the diagnostic value of liquid biopsy in assessing the molecular profile of non small cell carcinomas in comparison with tissue biopsy.
METHODS: We retracted eligible articles from PubMed, Embase and Cochrane databases. We calculated the pooled sensitivity (SEN), specificity (SPE), positive likelihood ratio (PLR), negative likelihood ratio (NLR) and diagnostic odds ratio (DOR). A summary receiver operating characteristic curve (SROC) and area under curve (AUC) were used to evaluate the overall diagnostic performance using the Meta-Disc software 5.1.32. The heterogeneity was assessed using I square statistics. A meta-regression was performed in case of heterogeneity. In case of absence of covariates, a sensitivity analysis was done in order to assess publications that induced a statistical bias.
RESULTS: 39 eligible studies involving 4782 patients were included. The overall statistical studies showed heterogeneity in the SEN, SPE, PLR, NLR and DOR. No threshold effect was revealed. The meta-regression incorporating the ethnicity, the test, the technique used in tissue and plasma and the use of plasma or serum as covariates showed no impact of these factors. A sensitivity analysis allowed achieving the homogeneity in the SPE and DOR. The overall pooled SEN and SPE were 0.61 and 0.95 respectively. The PLR was 9.51, the NLR was 0.45 and DOR was 24.58. The SROC curve with AUC of 0,93 indicated that the liquid biopsy is capable of identifying wild type samples from mutated ones with a relatively high accuracy.
CONCLUSION: This meta-analysis suggested that detection of molecular mutations by cfDNA is of adequate diagnostic accuracy in association to tissues. The high specificity and the moderate sensitivity highlight the value of LB as a screening test.

Background

Lung cancer is the leading cause of cancer-related death worldwide [1]. Its positive diagnosis is based on microscopic features and faced a recent change due to the 2015 World Health Organization Classification’s [1, 2]. For the first time, this classification introduced molecular pathways and targets especially for adenocarcinomas. In fact, this histologic subtype has become the most frequent non-small-cell lung carcinoma. This classification pointed out the necessity of 
not only assessing the accurate microscopic diagnosis but also the importance of molecular diagnosis of the most relevant targets. Lung cancer is mainly characterized by its spatial and temporal heterogeneity [3, 4]. Spatial heterogeneity consists in the presence in the same tumor of different molecular drivers. This fact compels to multiply samples in order to assess all the potential relevant pathways involved. On the other hand, temporal heterogeneity consists in the difference of activated pathways between the initial tumour and the metastases or the recurrences. This fact enhances the necessity of sampling the metastases or recurrences even if the initial tumoral profile was assessed. This heterogeneity provides also an explanation to the phenomenon of resistance, which is observed within 3 to 6 months after the onset of anti-EGFR treatments. This resistance is explained by the activation of secondary pathways that were activated at the onset but concerned a low number of tumour cells. The morphological and molecular tests are performed in 80% of the cases on small samples and molecular testing is impossible in 30% of the patients. This may be due to the unavailability of the specimen, the inaccessibility of the tumoral site or the presence of contraindications to biopsy [3]. This fact made scientists and researchers look for other surrogates to tissue that can be safer and sufficient to establish the molecular profile. In this context, the liquid biopsy was discovered. It consists in the assessment of molecular profile on circulating tumor cells, circulating tumoral DNA, circulating tumoral RNA, exosomes or secretomes [5]. Many studies were published concerning the assessment of these elements with varying techniques of identification. In 2014, the liquid biopsy was introduced in the management scheme of patient candidates for the third generation anti-EGFR in order to assess the presence of the T790M mutation [6, 7]. Besides, in 2016, the first technique of sequencing, the cobas EFGR mutation test, obtained the Food and Drug Administration approval [8, 9]. Even if this technique was approved, there are still many publications dealing with different techniques that may seem less expensive or easier to perform in a Pathology lab. Recently, many authors reported the efficiency of tests performed on free circulating DNA (cfDNA) in comparison with those performed on circulating tumour cells (CTC) [10]. We aimed to highlight the diagnostic value of liquid biopsy in assessing the molecular profile of non small cell carcinomas in comparison with tissue biopsy and we focused on the mutations of the Epidermal Growth Factor Receptor gene (EGFR). Other genes were assessed in only 4 included studies.

Methods

Data Source and Search

We conducted this meta-analysis under the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [11]. To retrieve all eligible articles, PubMed and Embase databases and Cochrane Library were comprehensively searched up to 01 June 2017 with limitation to French and English language. The search medical subject heading (MeSH) terms employed for literature retrieval included: lung cancer or lung neoplasm, cell free DNA or cfDNA or circulating DNA and diagnosis or sensitivity or specificity or accuracy. The reference list of eligible articles was also independently searched to obtain other valuable sources.

Study Selection Criteria

To be qualified for inclusion in this meta-analysis, articles must comply with all the following criteria: articles evaluated the diagnosis value of cfDNA in plasma/serum or blood for lung cancer, the diagnosis of lung cancer was confirmed by the gold standard test which is the biopsy and articles provided sufficient data (true negative (TN), true positive (TP), false negative (FN) and false positive (FP)). The major exclusion criteria were as follows: studies with duplicate data reported in other studies and reviews, technical reports, case reports, comments or letters with invalid data.

Data Extraction and Quality Assessment

One investigator independently reviewed all the articles and extracted data from the selected articles: first authors’ name, publication year, characteristics of participants (ethnicity, mean/median, age, source of control, number of cases and controls, sample types), assay methods, assay indicators, sensitivity, specificity and quality assessment information. In addition, based on the revised quality assessment of diagnosis, accuracy studies-2 (QUADAS-2) criteria, the included articles were evaluated as at high risk (H) or low risk (L) independently by four key domains: patient selection, index test, reference standard and flow and timing [12].

Statistical Analysis

We used the Meta-Disc software 5.1.32 to conduct this meta-analysis. The pooled sensitivity (SEN) (TP/TP+FN), specificity (SPE) (TN/TN+FP), negative likelihood ratio (NLR), positive likelihood ratios (PLR) and diagnostic odds ratio (DOR) with the 95% confidence intervals were calculated. At the same time, we constructed the summary receiver operator characteristic (SROC) curve and calculated the area under the SROC curve based on the SEN and SPE of each study.

Threshold Effect

A threshold effect was assessed using the Moses model with calculation of the Spearman correlation coefficient.

Heterogeneity

Q test and I2 statistics were carried out to explore the heterogeneity among studies. P value <0.1 for q test or I2 value >50% represented substantial between study heterogeneity. Besides, based on the characteristics of the included articles, meta-regressions were performed to explore the sources of heterogeneity if necessary.

Sensitivity Analysis

In case of absence of covariates, according to the meta-regression analysis, we performed a sensitivity analysis using the same software. This analysis is performed through a visual inspection of forest plots. Studies causing bias are those that show large deviations from the line corresponding to the pooled accuracy estimation mentioned in the forest plot of specificity. These studies were excluded and considered as possible sources of heterogeneity. The purpose of sensitivity analysis is to stipulate hypothesis about the sources of heterogeneity when metaregression shows no covariates.

RESULTS

Search Results

Our database research retrieved 839 records. After reviewing the title and abstracts, 729 records were excluded due to language limit, unrelated studies. By reviewing full-text articles, we excluded further 65 records, leaving 43 eligible articles and 2 international congress abstracts. From these articles, 16 records were excluded due to insufficient data (3 articles and 1 congress abstract), no gold standard (8 articles) and duplicate publications (4 articles). In the study reported by Li and colleagues, EGFR mutation was detected in both plasma and serum and the data of plasma and serum were analyzed as two independent studies [13]. Xu and coworkers described 3 different techniques for the specific analysis of the Exon19 deletion and the L858R mutation of the EGFR gene. So that, the different data were considered as 6 independent studies [14]. After independent review, 39 eligible studies were included in this meta-analysis. The 
Fig. ( illustrates the flow-chart of the literature retrieval.

Fig. (1)

Flow-chart illustrating the literature retrieval.

All the studies fulfilled the major QUADAS-2 categories with a global low risk of bias and low concerns concerning applicability. The quality assessment of the different studies included is represented in Table .

A total of 4,782 participants were included in the analysis. The majority of the patients presented a late stage lung cancer. All the studies dealt with the sequencing of the EGFR gene in association to the sequencing of TP53, NF1, KRAS, MET in 1 study [15], to BRAF in one study [16], to KRAS in 1 study [17] and 1 study dealt with the screening of the ALK gene [18]. The techniques of sequencing in the liquid biopsy and in the tissue were similar in 20 studies. In the other studies they were different. The molecular diagnosis was performed on liquid biopsy and tissue at the same time in 17 studies and was not specified in 10 studies. Many techniques of sequencing were used in liquid biopsy consisting in PCR-based-sequencing techniques and non PCR-based-sequencing techniques. PCR-based-sequencing techniques consisted in digital PCR (dPCR) [19], amplification refractory mutation system (ARMS) [17], CastPCR [16], peptide-nuclei-acid mediated PCR (PNA-PCR) [20], mutant-enriched PCR (ME-PCR) [21], High Resolution Melting (HRM) [22], mutant enriched-liquidchip PCR technique [14], PNA-LNA-PCR technique [23]. Non PCR-based tecniques consisted in next-generation sequencing (NGS) [18], Cobas EGFR mutation test [24], Therascreen [25] and denaturing high perforance liquid chromatography (DHPLC) technique [26]. The technique that was the most frequently used in this analysis was the scorpion ARMS technique. The NGS techniques were reported in only 5 studies. The Table summarizes the main characteristics of the included articles.

Diagnostic Accuracy of the Liquid Biopsy

The overall pooled SEN and SPE were 0.63 (95% CI, 0.61-0.65) and 0.92 (95% CI, 0.91-0.93) respectively (Figs. 2 and 3). Our results showed that PLR was 8.123 (95% CI, 5.13-12.84), NLR was 0.456 (95% CI, 0.383-0.543) and DOR was 20.50 (95% CI, 12.61-33.30) (Fig. ). Between-study heterogeneity was significant in the SEN, SPE and the DOR (I-square estimated to respectively 84.9%, 89.1% and 74.9%). We did not find any evidence of threshold effect (Spearman correlation coefficient: 0.029 and p=0.861). Fig. ( shows the corresponding SROC curve with AUC of 0,82 indicating that the liquid biopsy is capable of identifying wild type samples from mutated ones with a relatively high accuracy.

Fig. (5)

A) The summary operative receiver characteristic curve indicating the area under curve of all studies, B) Forrest plot of dOR after sensitivity analysis.

Subgroup Analysis

Sub-group analyses based on the use of the NGS technique, the use of scorpion ARMS technique, the use of DHPLC technique, the use of the same technique in the liquid biopsy and tissue and the analysis of specific mutations of the EGFR gene were also conducted. The NGS tehniques seem to have the highest sensitivity of 0.75 and the highest specificity was recorded in the group of the ARMS Scorpion technique. Even when we analyzed the group of studies using the same techniques in the tissue and the liquid

biopsy, we noticed a heterogeneity between the different studies. The studies screening the deletion in the exon19 and those reporting mutations of the different exons of the EGFR gene presented quite similar sensitivities and specificities with heterogeneity in all cases. Table illustrates the different results.

Heterogeneity and Meta-Regression Analysis

The meta-regressions were also performed to further explore potential sources of heterogeneity (Table ). Our meta-regression analysis characteristics included ‘ethnicity (Asian or not)’, ‘the technique (Next generation sequencing

or not)’, ‘tissue/plasma (same technique use in the tissue and plasma or not)’, ‘plasma/serum (studies performed on serum or not). We didn’t include the smoking status as a possible covariate because of its subjective estimation by the patients. The meta-regression results suggested that no covariates might be responsible for this heterogeneity.

Sensitivity analysis: A sensitivity analysis was performed because of the presence of heterogeneity with no covariates highlighted by the meta-regression. We focused on the specificity forest plot because liquid biopsy is considered as a diagnostic test and mustn’t induce the treatment of patients with no mutations. The sensitivity analysis excluded the studies of Que D, et al, Santos, et al, Douillard, et al, Huang, et al, Wang, et al, Weber, et al, Zhu, et al, Mack et al, Kuang et al, Sriram, et al, Xu, et al. and Sequist, et al. [30, 32, 34, 36, 38, 39, 40, 17, 36, 44, 14, 25]. The overall pooled SEN and SPE were 0.61 (95%CI, 0.58-0.64) and 0.95 (95%CI, 0.94-0.96) respectively (Figs. 2a and 2b). Our results showed that PLR was 9.51 (95%CI, 6.66-13.58), NLR was 0.45 (95%CI, 0.37-0.56) and DOR was 24.58 (95%CI, 15.23-39) (Figs. 3a, 3b and 4). We noticed no heterogeneity between studies in the SPE, PLR and the DOR (I-square estimated to respectively 42%, 33% and 43.9%). The area under curve was estimated to 0.93. We did not find any evidence of threshold effect (p=0.159).

Discussion

This meta-analysis highlighted the efficacy of liquid biopsy in determining the EGFR gene mutation status in non-small cell carcinoma. According to the suggested guidelines for interpretation of AUC, ctDNA had high accuracy (0.927]. In this meta-analysis, the majority of the studies were about late-staged carcinomas. Our sub-group analysis revealed a better sensitivity of next generation sequencing techniques with a better specificity of ARMS technique. Besides, even the stratified analysis of individual mutation, when applicable, showed relatively the same SEN and SPE with a significant heterogeneity between studies. Our sub-group analysis showed a heterogeneity even if studies were grouped based on the technique, the use of plasma or serum or the punctual mutation of the EGFR gene. The sensitivity analysis allowed achieving homogeneity by excluding 12 studies. The final group was characterized by the Asian ethnicity and the use of PCR-based techniques as diagnostic tests. It was quite surprising to exclude the study of Sequist et al. [25] which was based on NGS techniques. In their meta-analysis, Li and coworkers studied the importance of the country, the random or consecutive patient selection and test method and reported that test method was the unique contributing factor with p=0.00354 [28]. This meta-analysis included only 13 studies and the authors didn’t perform a subgroup analysis.

We would like to discuss the potential limitations of this work. The fact that we didn’t assess confounding factors highlights the multiplicity of these factors including the technical steps that are not discussed in the different studies, the percentage of tumor cells, the histologic subtype of the tumours, the collection time of blood sample, the detailed chemotherapy regimens that may be different sources of bias. Besides, most studies included tissue samples formalin-fixed paraffin-embedded which lead to significant DNA degradation and increase detection bias. This fact enhances further studies to investigate these issues.

Conclusion

This meta-analysis suggested that detection of molecular mutations by cfDNA is of adequate diagnostic accuracy in association with tissues. The high specificity and the moderate sensitivity highlight the value of liquid biopsy as a screening test.

Sources of funding

The authors report no external funding

Consent for Publication

Not applicable.

Table 1

The quality assessment of the different studies included.

Studiess			Risk of Bias				Applicability Concerns
			Patient Selection	Index Test	Reference Standard	Flow and Timing	Patient Selection	Index Test	Reference Standard
Kimura et al. 2007 [29]	2007	42	L	L	L	L	L	L	L
Bai et al. [26]	2009	230	L	L	L	L	L	L	L
Que et al. [30]	2016	121	L	L	L	L	L	L	L
Cui et al. [18]	2017	39	L	L	L	L	L	L	L
Rachiglio et al. [31]	2016	44	L	L	L	L	L	L	L
Santos et al. [32]	2016	63	L	L	L	L	L	L	L
Goto et al. [33]	2012	86	L	L	H	L	L	L	L
Douillard [34]	2013	652	L	L	H	L	L	L	L
He et al. [35]	2016	120	L	L	H	L	L	L	L
Yang et al.[16]	2017	107	L	L	H	L	L	L	L
Huang et al. [36]	2012	822	L	L	L	L	L	L	L
Liu et al. [10]	2013	86	L	L	L	L	L	L	L
Kim et al. [37]	2013	40	L	L	L	L	L	L	L
Zhao et al. [21]	2012	111	L	L	L	L	L	L	L
Wang et al. [38]	2014	134	L	L	L	L	L	L	L
Jing et al. [22]	2014	120	L	L	L	L	L	L	L
Weber et al. [39]	2014	196	L	L	L	L	L	L	L
Zhang et al. [19]	2016	215	L	L	L	L	L	L	L
Zhu et al. [40]	2015	172	L	L	L	L	L	L	L
Mack et al. [17]	2009	14	L	L	L	L	L	L	L
Kuang et al. [41]	2009	43	L	L	L	L	L	L	H
He et al. [42]	2009	18	L	L	L	L	L	L	L
Brevet et al. [29]	2011	31	L	L	L	L	L	L	L
Jiang et al. [43]	2011	58	L	L	H	L	L	L	L
Sriram et al. [44]	2011	64	L	L	L	L	L	L	L
Xu et al. [14]	2012	34	L	L	L	L	L	L	L
Xu et al. [14]	2012	34	L	L	L	L	L	L	L
Xu et al. [14]	2012	34	L	L	L	L	L	L	L
Xu et al. [14]	2012	34	L	L	L	L	L	L	L
Xu et al. [14]	2012	34	L	L	L	L	L	L	L
Xu et al. [14]	2012	34	L	L	L	L	L	L	L
Studiess			Risk of Bias				Applicability Concerns
			Patient Selection	Index Test	Reference Standard	Flow and Timing	Patient Selection	Index Test	Reference Standard
Kim et al. [20]	2013	57	L	L	L	L	L	L	L
Sequist et al. [25]	2015	227	L	L	L	L	L	L	L
Wu [45]	2015	24	L	L	L	L	L	L	L
Mok [24]	2015	238	L	L	L	L	L	L	L
Li [13]	2014	141	L	L	H	L	L	L	L
Li [13]	2014	108	L	L	L	L	L	L	L
HE [35]	2017	120	L	L	L	L	L	L	L
Yung et al.[29]	2009	35	L	L	L	L	L	L	L

Table 2

The major characteristics of the different studies included.

Study Year Number			TP	FP	FN	TN	Test	Genes	Ethnicity	Test of Biopsy	Time Point of Biopsy and Liquid Biopsy	Stage	Plasma/ Serum	CR
Yung et al. [29]	2009	35	11	0	1	23	Microfluidics digital PCR	Ex 19, L858R	Asian	Direct sequencing	No mention	No mention	plasma	97%
Kimura et al. 2007 [29]	2007	42	6	1	2	33	ARMS	Ex 18, 19, 21	Asian	Direct sequencing	BT liquid biopsy, not at the same time	III or IV	serum	92%
Bai et al. [26]	2009	230	63	16	14	137	DHPLC	Ex19, 21	Asian	DHPLC	No mention	IIIb ou IV	plasma	87%
Que et al. [30]	2016	121	34	10	10	67	DHPLC	Ex19, 21	Asian	ARMS	BT the same time	I-IIIa:17IIIb-IV:104.	plasma	83%
Cui et al. [18]	2017	39	13	0	11	15	NGS	ALK	Asian	NGS	Not at the same time	I-IIIa:7IIIb-IV:32	plasma	72%
Rachiglio et al. [31]	2016	44	17	2	5	20	NGS	EGFR	European	NGS	Not at the same time	IV	Plasma	84%
Santos et al. [32]	2016	63	33	10	15	5	NGS	EGFR, TP53, NF1, KRAS, MET	European	Not mentionned	Not specified	Not specified	plasma	60%
Goto et al. [33]	2012	86	22	0	29	35	Scorpion ARMS	EGFR	Asian	ARMS	Pre-TT both	Not specified	serum	66%
Douillard [34]	2013	652	69	1	36	546	Scorpion ARMS (ex19 del, L858R, T790M)	EGFR	European	Scorpion ARMS	Both BT.	Stage IIIa, b, IV	plasma	94%
He et al. [35]	2016	120	80	0	26	14	Targeted (ddPCR) (Ex19 del, L858R, T790M)	EGFR	Asian					78%
StudyYearNumber			TP	FP	FN	TN	Test	Genes	Ethnicity	Test of Biopsy	Time Point of Biopsy and Liquid Biopsy	Stage	Plasma/Serum	CR
Yang et al. [16]	2017	107	31	3	24	49	Cast PCR	EGFR, BRAF	Asian	Not mentionned	Not the same time	I-III:42IV:65	plasma	74%
Huang et al. [36]	2012	822	188	81	108	445	DHPLC	EGFR	Asian	DHPLC	THE SAME TIME	IIIb, IV: 744I-IIIa: 78	plasma	77%
Liu et al. [10]	2013	86	27	0	13	46	Scorpion ARMS	EGFR	Asian	ARMS	No mention	III ET IV	plasma	85%
Kim et al. [37]	2013	40	6	0	29	5 exclus	PNA-mediated real-time PCR	EX19 del, L858R	Asian	Direct sequencing	Not the same time	advanced	plasma	87%
Zhao et al. [21]	2012	111	16	3	29	63	ME-PCR(19 del, L858 R)	EGFR	Asian	ME-PCR	Same time BT	Not mentionned	plasma	71%
Wang et al. [38]	2014	134	15	0	53	64	(ARMS SCORPION)	EGFRVP: 15, FP: 2, VN: 4, FN: 53	Asian	ARMS	After TT	115 IV, 19 IIIb	plasma	59%
Jing et al. [22]	2014	120	29	2	16	73	HRM + direct sequencing	EGFR	Asian	HRM + direct sequencing	During surgery for liquid biopsy. Not at the same moment,	I-II: 38III-IV:82	plasma	85%
Weber et al. [39]	2014	196	17	6	11	162	NGS (cobas)	EGFR	European	cobas	BT liquid biopsy, not at the same time	I, II:2III, IV: 197	plasma	91%
Zhang et al. [19]	2016	215	57	4	36	118	ddPCR	Ex19 del, L858R	Asian	ARMS	The same, AT	IIIb:36IV:179	plasma	81%
Zhu et al. [40]	2015	172	30	4	7	131	Targeted (ddPCR)	Ex19 del, L858R	Asian	ARMS	No mention	Not mention	plasma	93%
Mack et al. [17]	2009	14	4	4	2	4	(scorpion ARMS)	EGFR, KRAS	American	Nested PCR assay	BT liquid biopsy not mentionned for tissue	IIIb et IV	plasma	57%
Kuang et al. [41]	2009	43	21	9	2	11	Scorpion ARMS	Ex18, 19, 20	American	Direct DNA sequencing or DNA endonuclease-based method (local)	AT liquid biopsy not the same time as biopsy	III or IV	plasma	74%
He et al. [42]	2009	18	8	0	1	9	ME-PCR	Ex19del, Ex 21 L858R	Asian	Direct sequencing	BT liquid biopsy, not at the same time.	Not specified	plasma	94%
StudyYearNumber			TP	FP	FN	TN	Test	Genes	Ethnicity	Test of Biopsy	Time Point of Biopsy and Liquid Biopsy	Stage	Plasma/Serum	CR
Brevet et al. [29]	2011	31	7	2	11	11	Mass spectrometry genotyping assaya	Ex19 del et Ex21 L858R	American	PCR-RFLP	BT liquid biopsy not always at the same time.	III or IV	plasma	58%
Jiang et al. [43]	2011	58	14	0	4	40	ME-PCR	Ex19, 21	Asian	Not specified	BT the same time	IIIb, IV	serum	93%
Sriram et al. [44]	2011	64	3	0	3	58	ME-PCR and HRM	EGFR: Ex19 et 21	European	ME-PCR et HRM	THE SAME TIME	Not specified	serum	95%
Xu et al. [14]	2012	34	4	4	4	23	ARMS	EGFR 19 del	Asian	ARMS	Liquid AT and tissue BT	IIIb a IV	Plasma	79%
Xu et al. [14]	2012	34	4	0	4	26	ARMS	EGFR L8585R	Asian	ARMS	Liquid AT and tissue BT	IIIb a IV	Plasma	88%
Xu et al. [14]	2012	34	0	1	7	26	DHPLC	EGFR 19 del	Asian	ARMS	Liquid AT and tissue BT	IIIb a IV	Plasma	76%
Xu et al. [14]	2012	34	2	2	6	24	DHPLC	EGFR L8585R	Asian	ARMS	Liquid AT and tissue BT	IIIb a IV	Plasma	76%
Xu et al. [14]	2012	34	2	5	5	22	ME-liquidchip	EGFR 19 del	Asian	ARMS	Liquid AT and tissue BT	IIIb a IV	Plasma	70%
Xu et al. [14]	2012	34	2	1	6	25	ME-liquidchip	EGFR L8585R	Asian	ARMS	Liquid AT and tissue BT	IIIb a IV	Plasma	79%
Kim et al. [20]	2013	57	8	3	4	42	PNA-LNA PCR (EGFR), sequencing (KRAS)	EGFR, KRAS	Asian	Direct sequencing	The same time BT	IIIb, IV	serum	87%
Sequist et al. [25]	2015	227	155	23	37	12	NGS (cobas or therascreen)	EGFR	American	Cobas or therascreen	The same time	IV after progreesion	plasma	73%
Wu Ya-Lan [45]	2015	24	7	2	10	5	ARMS	EGFRT 790M	Asian	ARMS	Yes after treatment	IV	plasma	50%
Mok [24]	2015	238	72	5	24	137	cobas	EGFR	Asian	Cobas	Yes before TT	IIIb, IV	plasma	87%
Li [13]	2014	141	27	3	29	62	ARMS	EGFR 19 del, L858R, T790M	Asian	ARMS	Not specified	IIIb, IV	plasma	63%
Li [13]	2014	108	19	2	29	42	ARMS	EGFR 19 del, L858R, T790M	Asian	ARMS	Not specified	IIIb, IV	serum	56%
HE [35]	2017	120	80	0	26	14	ddPCR	EGFR, Ex19del, L858R, T790M	Asian	ddPCR	At the same time, BT	Advanced stage	plasma	78%

CR: concordance rate.

Table 3

The pooled sensitivities, specificities and I-square of the sub-groups: NGS technique, ARMS technique, DHPLC technique, same technique in tissue and liquid plasma, screening of Ex19 deletion and L858R mutation, screening of Exons 18, 19, 20.

Sub-groups	Pooled-SEN	Pooled-SPE
NGS
Cui S et al.Rachiglio et al.Santos et al.Sequist et al.Mok et al	0.75 [0.71-0.801]I2: 56.9%	0.82 [o.77-0.87]I2:95.6%
ARMS
Goto et al.Douillard et al.Liu et al.Wang et al.Mack et al.Kuang et al.Xu et al.Xu et al.Wu et al.Li et al.Li et al	0.509 [0.461-0.558]I2: 83.5%	0.972 [0.95-0.98]I2:90.3%
DHPLC
Bai et al.Que et al.Huang et al.Xu et al.Xu et al	0.66 [0.618-0.709]I2: 88.1%	0.86 [0.83-0.88]I2:40.6%
Same technique Tissue/Biopsy
Bai et al.Cui et al.Rachiglio et al.Goto et al.Douillard et al.Huang et al.Liu et al.Zhao et al.Wang et al.Jing et al.Weber et al.Sriram et al.Xu et al.Xu et al.Sequist et al.Wu et al.Mok et al.Li et al.Li et al.He et al	0.63 [0.6-0.65]I2:87.3%	0.93 [0.91-0.94]I2:92.4%
Sub-groups	Pooled-SEN	Pooled-SPE
Ex19 del and L858R mutation
Yung et alBai et alQue et alKim et alZhang et alZhu et alHe et alBrevet et alJiang et alSriram et al	0.66 [0.61-0.71]I2:86.6%	0.94 [0.92-0.96]I2:72%
Screening exons 18, 19, 20
Kuang et alKimura et alLi et alLi et alHe et al	0.63 [0.57-0.69]I2:88.1%	0.91 [0.86-0.95]I2:84.4%

Table 4

Meta-regression analyzing 3 covariates: the test (NGS or not), the ethnicity (Asian or not), the test used in the tissue and the liquid biopsy (the same or not), the use of plasma or serum (serum or not).

Covariates	Coefficients	P value
Ethnicity	0.01	0.98
Test	-0.49	0.4
Test tissue versus liquid biopsy	0.45	0.37
Plasma versus serum	0.48	0.53