Background

Lung cancer was the second most commonly diagnosed cancer and the leading cause of cancer death in 2020, with an estimated 2.2 million new cancer cases and 1.8 million deaths worldwide [1]. Approximately 60% of lung cancer patients have distant metastasis at the initial diagnosis, and a substantial number of patients still progress to local recurrence or distant metastasis after curative-intent treatment [2, 3]. Low-dose computed tomography (LDCT) has been suggested for lung cancer screening, with a significant relative reduction in mortality for lung cancer patients [4]. However, distinguishing small malignant nodules in LDCT from benign lesions is particularly challenging due to the high false-positive rate [5,6,7]. In practice, a significant proportion of patients still suffer from ambiguous disease progression during radiological follow-up after radical surgery. Presently, the prognostic stratification of non-small cell lung cancer (NSCLC) using factors such as tumor, node, metastasis (TNM) classification, airway spread, and pathological subtype has shown limited effectiveness, with some low-risk patients experiencing postoperative relapse [8]. Therefore, more effective approaches and biomarkers will contribute to early detection, precision medicine, individualized treatment, and prognostication of lung cancer, and they are urgently needed.

Recently, with the advantages of noninvasiveness, ease of access, reproducibility, good reflection of the overall state of the tumor, and real-time surveillance, liquid biopsy has been widely researched and has shown promising efficacy in early diagnosis, prognostication, and disease monitoring in lung cancer. Circulating tumor cells (CTCs), circulating tumor DNA (ctDNA), methylation signatures, and microRNAs are the most commonly detected biomarkers [9, 10]. The clinical applications of liquid biopsy have been gradually established for advanced-stage NSCLC and metastatic disease. With the development of molecular biological detection technologies and platforms, combined with multianalytical approaches and machine learning models, the clinical performance of liquid biopsy in detecting diagnostic, prognostic, and predictive biomarkers for early-stage NSCLC has been further investigated in the last few decades [11,12,13,14,15,16]. Here, we performed this systematic review and meta-analysis to discuss the advantages and current limitations of liquid biopsy in the management of localized NSCLC, regarding the commonly detected biomarkers, CTCs, ctDNA, methylation signatures, and microRNAs. The potential clinical utility of liquid biopsy in early-stage NSCLC diagnosis, prognosis, and clinical monitoring of treatment response or recurrence will also be explored.

Methods

Study selection

This study followed the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) 2020 statements checklist (Additional file 1: Table S1) [17]. Literature search was conducted in Cochrane Library, PubMed, EMBASE databases, ClinicalTrials.gov, and reference lists by 2 researchers (H.S. and Y.J.) independently. Studies published since inception to 17 May 2022 were included. To perform a comprehensive search, we used the following keywords and MeSH terms in different patterns: (“Lung Neoplasm”) AND (“Liquid Biopsy” OR “Circulating Tumor Cell” OR “Circulating Tumor DNA” OR “DNA Methylation” OR “Circulating Tumor RNA”) (Additional file 2: Table S1).

Eligibility criteria

The following criteria were used for study inclusion: studies that evaluated the diagnostic and prognostic values of liquid biopsy for early-stage NSCLC, regarding the common biomarkers, CTCs, ctDNA, methylation signatures, and microRNAs; adequate data to construct the diagnostic 2 × 2 table for diagnostic assessment; sufficient survival data to obtain the hazard ratios (HRs) for the prognostic analysis in the preoperative, postoperative, and postchemotherapy time point; and the most recent or completed study if based on overlapping patients. The exclusion criteria were as follows: studies without any relevant data for analysis; stage IV or advanced-stage NSCLC; the involved sample size was fewer than 10; papers that were not published in English; and commentaries, editorials, reports, reviews, letters, and experiments.

Data extraction

The following data were extracted in a standardized form: the publication details, study design, patient characteristics, stages, biopsy method, type of biomarker, the true positive, false positive, false negative, true negative for the analysis of the sensitivity and specificity. Additionally, the concentrations of circulating cell-free DNA (cfDNA) in preoperative plasma of eligible cohorts were also extracted for the exploration of related clinical factors, including sex, age, TNM stage, smoking, and histopathology. And if the studies reported the association between circulating biomarkers and short- and long-term outcomes of the NSCLC patients, follow-up duration and the hazard ratio (HR) with a 95% confidence interval (CI) were extracted from the regression or survival plot of recurrence-free survival (RFS) and overall survival (OS). The prognostic analysis for molecular residual disease (MRD) detection focused on the survival data at postoperative time point. The reported lead time of biomarker (e.g., ctDNA) detection preceding radiographic progression was also listed and summarized. As for predictive value analysis, original survival data were extracted from the Kaplan–Meier curves comparing RFS between MRD/ctDNA-positive patients receiving adjuvant therapy and not receiving adjuvant therapy and comparing RFS between MRD/ctDNA-negative patients receiving adjuvant therapy and not receiving adjuvant therapy. In terms of treatment response evaluation for neoadjuvant therapy, relevant data in ctDNA clearance and assessment of pathological response were collected and analyzed. Any discrepancies were assessed by a third author (K.C.).

Risk of bias assessment

The quality assessment of the included studies was evaluated by the Quality Assessment for Studies of Diagnostic Accuracy Score-2 (QUADAS-2) tool [18], with 4 different domains: patient selection, index test, reference standard, and flow and timing. And for the cohort studies involved in prognostic analysis, the quality assessment was followed by the Newcastle–Ottawa Scale [19]. Publication bias was detected by Deeks’ funnel plot.

Statistical analysis

Diagnostic meta-analysis was based on the MIDAS module (Stata module for meta-analytical integration of diagnostic test accuracy studies) [20] and bivariate approach [21]. The pooled sensitivity and specificity were calculated by the accuracy data, and the summary receiver operative curve (SROC) was generated by “mada” package, and the area under the curve (AUC) was evaluated. Analysis of factors related to preoperative cfDNA concentration was conducted by Kruskal–Wallis test. In terms of prognostic analysis, the RFS and OS were measured by HRs and 95% CIs that were directly reported in the included studies. Otherwise, survival data that were not presented numerically in articles were extracted from the Kaplan–Meier curve using Engauge Digitizer version 12, and the HRs were calculated by the Parmar and Tierney methods [22, 23]. The random-effects model was pooled due to the high heterogeneity of the studies (p < 0.10 or I2 > 50%). Otherwise, the fixed-effects model was used. Subgroup analysis was performed by the preoperative, postoperative, and postchemotherapy time point, while the prognostic analysis for MRD detection was based on the survival data at postoperative time point. The lead time analysis was estimated and summarized by the Wan and Luo methods [24, 25] and then pooled and presented in an estimated average using the ggplot2 package. The forest plots of predictive value analysis were pooled in MRD/ctDNA-positive group and MRD/ctDNA-negative group. Correlation between pathological response and ctDNA clearance was conducted by the 2-sided Fisher’s exact test [26]. Survival analysis in pathological response and ctDNA clearance was also pooled in a forest plot. Statistical analyses were performed by Stata 15.0 software (Stata Corporation, College Station, TX) and RStudio 4.1.3. Statistical significance was set at p < 0.05.

Results

Literature selection and study characteristics

A flow diagram of the literature search is shown in Additional file 3: Fig. S1. After exclusion of studies, a total of 53 articles were included in the meta-analysis with 34 studies [14, 16, 27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58] eligible for diagnostic assessment and 21 [12,13,14,15, 33, 59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74] for prognostic analysis. Summary characteristics of the included studies were demonstrated in Additional file 2: Table S2.

Assessment of bias

Assessment of the study quality was evaluated using the QUADAS-2 tool (Additional file 2: Table S3) and Newcastle–Ottawa Scale (Additional file 2: Table S4). There was no significant publication bias determined by the Deeks’ funnel plot (p = 0.39, Additional file 3: Fig. S2) in diagnostic study nor by formal statistical tests of Egger’s test in prognostic analysis (Additional file 3: Fig. S3).

Diagnostic performance

For the diagnostic analysis, 2917 healthy controls and 3015 patients with early-stage NSCLC were included, among whom 1537 patients were limited to stage I. The biomarkers included CTC in 6 eligible studies with 885 participants, ctDNA in 7 studies with 1001 participants, DNA methylation in 11 studies with 1888 participants, and microRNAs in 7 relevant articles with 1079 participants. The estimated diagnostic values of different common biomarkers for early-stage NSCLC with AUCs ranged from 0.84 to 0.87 (Fig. 1A). Additionally, the comparison of the diagnostic values between different biomarkers showed no significant differences, with a similar ROC-AUC and overlapping 95% confidence ellipses. In particular, a lower AUC was calculated when the analysis was limited to stage I disease (Additional file 2: Table S5).

Fig. 1
figure 1

A Comparison of SROC of different biomarkers. B Analysis of cell-free DNA concentration

Full size image

Analysis of the cfDNA concentration

Eight cohorts of 941 patients with early-stage NSCLC were reanalyzed [13,14,15, 35, 40, 42, 66, 69]. The relative factors included TNM stage (I, II, and III), T stage (T1, T2, T3, and T4), N stage (N0, N1, and N2), age (≥ 65 vs. < 65), sex, smoking status, and histopathology. The median cfDNA concentration of the preoperative plasma samples from all patients with stage I–III NSCLC was 8.64 ng/ml (Additional file 2: Table S6, original data in Additional file 4: Table S1). For TNM stage, the median cfDNA concentration in stage I was 7.58 ng/ml, significantly lower than the median concentrations in stage II (9.86 ng/ml, p < 0.001) and III (10.03 ng/ml, p < 0.0001). There was no significant difference between the median concentrations in stages II and III (p = 0.22) (Fig. 1B). The median cfDNA concentration in the T1 stage was 1.09 ng/ml, significantly lower than the median concentrations in the higher T stages. Similarly, the median cfDNA concentrations in N0 stage (1.70 ng/ml) were significantly lower than in N1 and N2 stages. Older patients had significantly higher cfDNA concentrations than younger patients (≥ 65 vs. < 65, 9.03 ng/ml vs. 6.99 ng/ml, p < 0.000001). Smoking was also associated with significantly higher cfDNA concentrations (smoking vs. nonsmoking, 10.64 ng/ml vs. 7.52 ng/ml, p < 0.000001). No significant difference was detected between the sexes (female vs. male, 7.83 ng/ml vs. 7.90 ng/ml, p = 0.98). For different histopathologies, the results showed that lung adenocarcinoma (LUAD) was associated with significantly lower cfDNA concentrations than lung squamous cell carcinoma (LUSC) (LUAD vs. LUSC, 9.03 ng/ml vs. 10.78 ng/ml, p = 0.014) (Fig. 1B).

Prognostic performance

To evaluate the prognostic values of liquid biopsy in early-stage NSCLC, 21 eligible studies with 2143 patients were included [12,13,14,15, 33, 59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74]. Most of the studies referred to ctDNA biomarker, with only one relevant study focusing on DNA methylation [33], 2 regarding CTCs [60, 61], and 1 referring to microRNAs [59]. Subgroup analysis was performed by the preoperative, postoperative, and postchemotherapy time point, while the prognostic analysis for MRD detection was based on the survival data at postoperative time point. The forest plots showed that MRD detection after curative intent treatment was a strong predictor of disease relapse (RFS, HR, 4.95; 95% CI, 3.06–8.02; p < 0.001, Fig. 2A) and a shorter OS (HR, 3.93; 95% CI, 1.97–7.83; p < 0.001, Fig. 2B). Likewise, biomarkers positive in the preoperative blood samples were also associated with significantly inferior RFS (HR, 3.00; 95% CI, 2.12–4.24; p < 0.001, Fig. 2A) and OS (HR, 3.65; 95% CI, 1.96–6.77; p < 0.001, Fig. 2B). Similar trends were detected at the postchemotherapy time point for both RFS (HR, 4.51; 95% CI, 2.27–8.94; p < 0.001, Fig. 2A) and OS (HR, 4.41; 95% CI, 0.52–37.29; p = 0.17, Fig. 2B).

Fig. 2
figure 2

Forest plots of A recurrence-free survival and B overall survival. C Summary of lead time analysis

Full size image

Lead time analysis

A summary of the lead time by biomarker detection preceding radiographic progression was presented in Fig. 2C. All of the 9 eligible studies referred to ctDNA detection, with a total of 148 patients suffering disease progression [12, 13, 15, 68, 69, 71, 72, 74, 75]. Molecular recurrence was detected by next-generation sequencing (NGS) approaches and cancer personalized profiling by deep sequencing (CAPP-Seq), circulating single-molecule amplification and resequencing technology (cSMART), and multiplex polymerase chain reaction NGS (mPCR-NGS) platforms. The reported lead time ranged from 0 to 795 days. Taken together, our study found an average lead time of 179 ± 74 days.

Predictive value

The detailed data for predictive value analysis of MRD detection in adjuvant therapy guidance after lung cancer resection are presented in Fig. 3A [13, 66, 69, 72, 75, 76]. All of the 6 eligible cohorts explored the ctDNA-based MRD predictive value for adjuvant therapy (Additional file 4: Table S2). In this series, adjuvant therapy was found to confer a survival benefit for patients with detectable ctDNA-based MRD, while adjuvant therapy could not improve survival for undetectable MRD patients. Furthermore, the forest plot showed that the application of adjuvant therapy significantly benefited long-term survival in patients with ctDNA-based MRD + (RFS, HR, 0.27; 95% CI, 0.17–0.44; p < 0.001), while an opposite tendency was detected for MRD − patients (RFS, HR, 1.51; 95% CI, 0.81–2.79; p = 0.19) (Fig. 3B).

Fig. 3
figure 3

Analysis of predictive value and treatment response evaluation

Full size image

Treatment response evaluation

There were 3 studies eligible for the analysis of correlation between pathological response and ctDNA responder for neoadjuvant therapy (Additional file 2: Table S7) [71, 77, 78]. The percentage of patients with a major pathological response (MPR) or pathologic complete response (pCR) was higher among those with ctDNA responder (33% ~ 86%) than among those without ctDNA responder (0 ~ 17%) (Fig. 3C). Further Fisher’s exact test indicated a strong correlation between a pathological response and ctDNA responder (p < 0.00001). In addition, significantly improved long-term survivals were both observed for patients with pCR (HR, 0.41, 95% CI, 0.21–0.83, p < 0.05) and ctDNA clearance (HR, 0.16, 95% CI, 0.07–0.37, p < 0.001) after neoadjuvant therapy (Fig. 3D) [73, 77, 78].

Discussion

The study is a comprehensive systematic review and meta-analysis to evaluate the clinical utility of liquid biopsy focusing on the early-stage NSCLC, regarding the commonly detected biomarkers, CTCs, ctDNA, methylation signatures, and microRNAs. The present study indicated liquid biopsy could reliably facilitate more precision and effective management of early-stage NSCLC.

The early detection of lung cancer is critical for reducing its mortality and morbidity. In recent decades, many studies have investigated the screening and diagnostic values of different biomarkers, including ctDNA, CTCs, DNA methylation, and microRNAs. Our analysis indicated that these common biomarkers could provide similar diagnostic accuracy for early-stage NSCLC, with acceptable effectiveness in AUCs ranging from 0.84 to 0.87. With the development of platforms and technology, ctDNA analysis has provided a practical approach to the noninvasive detection of early-stage tumors [14, 35, 39]. In addition, the analysis of cfDNA fragmentomics profiles was newly highlighted as the novel approach and pattern for the accurate screening, early detection, and monitoring of human cancer [79, 80]. With the advantages of an early appearance in the disease course, cancer specificity, biological stability, and ready accessibility in bodily fluids [81], aberrant DNA methylation analysis combined with machine learning provided the opportunity to overcome the challenge of the low abundance of ctDNA in plasma samples [28, 29, 34, 58, 82, 83]. CTCs in the blood could serve as effective screening and diagnostic markers to discriminate small malignant pulmonary nodules from benign lesions [54, 56]. The profiling of circulating microRNAs has promising accuracy for discriminating lung cancer patients from healthy controls and could be developed as a supplement in future screening [51, 84]. Recently, integrative multi-analytical models combining clinical features and multiple biomarkers have achieved a better balance of predictive sensitivity and specificity [16, 40, 53]. Multiomics analysis combining ctDNA detection with methylation, exosomes, circulating microRNAs, circulating tumor cells, metabonomics, and molecular imaging methods should be considered to improve the efficacy and provide more practical value for clinical applications.

In this study, there was a lower diagnostic accuracy for stage I NSCLC. The low concentration of cfDNA molecules in plasma introduces a biological limitation for detecting early-stage tumors [85, 86]. Similar to the previous studies [85, 87], our analysis pooled a median cfDNA concentration of 8.64 ng/ml in early-stage NSCLC and suggested that TNM stage I, T1 stage, N0 stage, adenocarcinoma, young age, and nonsmoking were associated with significantly lower cfDNA concentrations. The summarized data on the limited tumor burden highlighted the physical limitation of ctDNA analyses, particularly for the early detection of NSCLC patients with T1a-c stage. Additionally, clonal hematopoiesis of undetermined potential (CHIP) during aging is a common confounding factor influencing ctDNA detection. Deep sequencing of both white blood cell DNA and cfDNA might be required to identify and filter out CHIP-related mutations to reduce false-positive ctDNA detection rates [85].

With the development of new technologies and platforms, the feasibility and efficacy of MRD assessment and postoperative disease monitoring for early-stage NSCLC have gradually been investigated [85, 88, 89]. It is now widely recognized that patients with positive ctDNA and MRD detection after curative intent treatment have a worse prognosis than those with undetectable MRD [12, 15, 66, 72, 75]. In addition, the promising ctDNA MRD analysis for the early detection of disease recurrence preceding radiographic progression by an average lead time of 179 ± 74 days could reliably facilitate more effective interventions, offering the chance for earlier treatment strategy decision-making and to treat patients at their lowest tumor burden [12, 13, 15, 71, 74]. Although the other biomarkers including CTC, DNA methylation and microRNAs also showed the promising prognostic values, there is still lack of clinical cohort regarding their MRD detection during post-treatment monitoring [59, 60, 84].

A summary of the four different MRD detection patterns is presented in Table 1. For assay design, a tumor-naive panel involves sequencing several genes commonly known to be mutated in lung cancer, and thus there is no need for individualized tumor information [90]. The novel CAPP-Seq strategy has developed and achieved the enhancements in ctDNA analysis, with the ultrasensitive detection efficiency under the ultralow detection limit [14, 91]. Correspondingly, tumor-informed assays were designed based on the whole exome/genome sequencing of tumor tissues and tumor-adjacent normal tissues. The personalized panels take advantage of accurately tracking a larger number of mutations for each patient and improving the sensitivity of ctDNA detection [92]. However, the method is more expensive and has limitations in detecting de novo resistance/oncogenic alterations. The novel MRDetect model demonstrated that increased breadth or an expanded number of targeted mutations via genome wide mutational integration could effectively overcome the limitation of cfDNA abundance with a modest sequencing depth [11]. Honestly, DNA methylation analysis was proved to provide the promising efficacy in cancer detection and screening [82, 83], but there is still a lack of large-scale validations and clinical implementations regarding their MRD detection during post-treatment monitoring in early-stage lung cancer [93]. The currently ongoing prospective cohort, which was designed to investigate the feasibility of tumor-informed methylation-based MRD detection and postoperative cancer surveillance, is fully expected [94].

Table 1 Summary of MRD detection patterns
Full size table

Increasing evidence indicates that ctDNA-based MRD detection could provide good prognostic value for early-stage NSCLC after surgical resection and adjuvant therapy. However, evidence of the predictive value of MRD detection for adjuvant therapy guidance is still lacking. Our results indicated that the application of adjuvant therapy significantly benefited long-term survival in patients with ctDNA-based MRD + , while adjuvant therapy could not improve survival for MRD − patients [13, 66, 69, 72, 75, 76]. These results clearly revealed that longitudinal MRD monitoring could provide clinical utility for individualized patient care with adjuvant therapies and could efficiently avoid overtreatment of low-risk patients. However, the evidence should be interpreted cautiously because of the small sample sizes and lack of prospective interventional study designs. Thus, randomized controlled trials are necessary to confirm the predictive value of MRD. The MERMAID-1 trial [95] is ongoing to assess the efficacy of adjuvant durvalumab combined with chemotherapy in postsurgical MRD + patients with stage II–III NSCLC. Meanwhile, the ongoing MERMAID-2 trial [96] is designed to evaluate the efficacy and safety of durvalumab adjuvant therapy in stage II–III NSCLC patients who become MRD + during the surveillance period after curative intent therapy. The results of these relevant trials are highly anticipated, and high-level evidence is urgently awaited (Table 2) [97,98,99].

Table 2 Summary of the ongoing trails
Full size table

Recently, the clinical utility of ctDNA in predicting the response to neoadjuvant treatments and assessing the prognosis of NSCLC has gradually been explored [71, 77, 100]. The NADIM trial [73] reported for the first time a significant association between ctDNA levels after neoadjuvant chemoimmunotherapy and survival outcomes in operable NSCLC. The data were supported by 3-year OS and revealed that ctDNA outperformed radiologic assessments in the prediction of survival, which highlighted the usefulness of ctDNA as an early surrogate end point for neoadjuvant treatment. Likewise, CheckMate 816 [78] showed that ctDNA clearance was associated with a pathologic response, and event-free survival appeared longer in patients with ctDNA clearance than in those without, suggesting that clearance during neoadjuvant therapy may be an early predictor of favorable outcomes. Correspondingly, our study detected a strong correlation between pathological response and ctDNA clearance, and significantly improved long-term survivals were both observed for patients with pCR and ctDNA clearance after neoadjuvant therapy. These results revealed the effective utility of ctDNA status and dynamics analysis and provided a landmark approach for evaluating the response of therapeutic outcomes for resectable NSCLC patients treated with neoadjuvant therapy. We require more evidence and data regarding the treatment response evaluation of ctDNA analysis in lung cancer disease monitoring.

Limitations

Several limitations of this study should be considered. First, the types of liquid biomarkers, detection technologies, and platforms and the lack of a standardized ctDNA detection manual all contribute to the great heterogeneity among the included studies. Second, there were 10 extensive stage patients enrolled in our analysis [16, 63, 69], but their impact may be negligible. In addition, even though we have completed a comprehensive and systematic search of the literature, publication bias is still inevitable. Finally, some studies did not report detailed information on histology and radiology, and further analysis was therefore limited.

Conclusions

In conclusion, our study indicated liquid biopsy could reliably facilitate more precision and effective management of early-stage NSCLC, regarding the commonly detected biomarkers, CTCs, ctDNA, methylation signatures, and microRNAs. The lead time analysis suggested ctDNA detection monitoring could provide the opportunity for earlier interventions during disease surveillance. Improvement of liquid biopsy techniques and detection approaches and platforms are still needed, and higher-quality trails are required to provide more rigorous evidence prior to their routine clinical application.