Using Circulating Tumor DNA as a Novel Biomarker to Screen and Diagnose Colorectal Cancer: A Meta-Analysis

(1) Background: Circulating tumor DNA (ctDNA) has emerged as a promising biomarker for many kinds of tumors. However, whether ctDNA could be an accurate diagnostic biomarker in colorectal cancer (CRC) remains to be clarified. The aim of this study was to evaluate the diagnostic accuracy of ctDNA in CRC. (2) Methods: PubMed, Web of Science, and Cochrane databases were searched to identify studies reporting the use of ctDNA to screen and diagnose CRC, and all relevant studies published until October 2022 were enrolled for our analysis. These studies were divided into three primer subgroups: the subgroup of quantitative or qualitative analysis of ctDNA and the subgroup of septin9 (SEPT9) methylation assay. (3) Results: A total of 79 qualified articles with 25,240 subjects were incorporated into our meta-analysis. For quantitative studies, the combined sensitivity (SEN), specificity (SPE), and diagnostic odds ratio (DOR) were 0.723 (95% CI: 0.623–0.803), 0.920 (95% CI: 0.827–0.966), and 23.305 (95% CI: 9.378–57.906), respectively, yielding an AUC of 0.860. The corresponding values for qualitative studies were 0.610 (95% CI: 0.566–0.651), 0.891 (95% CI: 0.878–0.909), 12.569 (95% CI: 9.969–15.848), and 0.823, respectively. Detection of SEPT9 methylation depicted an AUC of 0.879, with an SEN of 0.679 (95% CI: 0.622–0.732), an SPE of 0.903 (95% CI: 0.878–0.923), and a DOR of 20.121 (95% CI:14.404–28.106), respectively. (4) Conclusion: Blood-based ctDNA assay would be a potential novel biomarker for CRC screening and diagnosis. Specifically, quantitative analysis of ctDNA or qualitative analysis of SEPT9 methylation exhibited satisfying diagnostic efficiency. Larger sample studies are needed to further confirm our conclusions and to make the ctDNA approach more sensitive and specific.


Introduction
Colorectal cancer (CRC) is the third most commonly diagnosed cancer and the second most common cause of cancer death worldwide, which accounts for nearly 10% of cancer cases and 9% of cancer deaths [1]. Even though there have been dramatic advances in understanding the molecular mechanism and clinical treatment over decades, helping to double the overall survival of advanced CRC to three years, there was estimated over 0.9 million CRC-related death in 2020 [2]. Since there are significant differences in prognosis and therapy response between different clinical stages in CRC patients, it is extremely important to diagnose and treat CRC at an early stage [3]. However, most CRC patients are asymptomatic at an early stage. When common symptoms such as constipation, bleeding, anemia, or abdominal pain are noted, CRC may already develop to advanced stages [4]. Currently, several methods, including fecal occult blood test (FOBT), fecal immunochemical test (FIT), and colonoscopy, have been applied in CRC screening, which significantly reduced CRC mortality [5]. However, there are some defects in the current screening method. The sensitivity (SEN) and specificity (SPE) of FOBT and FIT for early CRC are unsatisfactory [6,7]. Additionally, the FIT demonstrated poorer compliance in CRC monitoring

Quality Assessment
The quality assessment was based on QUADAS-2, and the outcome of eligible studies is depicted in Figures 2 and S1. The majority of incorporated studies possessed a moderatehigh quality, indicating that the overall quality of the included studies was highly acceptable. However, there were some concerns about the patient selection of 39 studies, such as whether the patient selection was consecutive or random in a specific time period. Moreover, predefined threshold information was absent in 22 studies, which potentially generated some concerns about the index test. All enrolled patients obtained definite pathological diagnoses. J. Clin. Med. 2023, 11, x FOR PEER REVIEW 5 of 18

Quality Assessment
The quality assessment was based on QUADAS-2, and the outcome of eligible studies is depicted in Figures 2 and S1. The majority of incorporated studies possessed a moderate-high quality, indicating that the overall quality of the included studies was highly acceptable. However, there were some concerns about the patient selection of 39 studies, such as whether the patient selection was consecutive or random in a specific time period. Moreover, predefined threshold information was absent in 22 studies, which potentially generated some concerns about the index test. All enrolled patients obtained definite pathological diagnoses.
(a) (b) Figure 6. Forest plots of SEN (a) and SPE (b) for diagnostic value of ctDNA assay for CRC in the SEPT-9 methylation detection subgroup. 95% CI, 95% confidence interval; CRC, colorectal cancer; ctDNA, circulating tumor DNA; SEN, sensitivity; SPE, specificity; diamond shapes representing average effect; dark blue squares representing individual study results.

Subgroup Analysis and Meta-Regression Analysis
Different covariates including region (Asia vs. non-Asia), sample size (≥100 vs. <100), control type (health control vs. non-CRC disease), sample source (plasma vs. serum), assay methods (RT-PCR vs. other methods in quantitative analysis; MSP vs. other methods in qualitative analysis), and methylation gene (SEPT9 vs. other genes in qualitative studies) were applied in the subgroup analysis (Table 2).
Subgroup analysis based on sample size revealed an improved diagnostic accuracy in quantitative studies with sample size ≥ 100 since the SEP, SEN, and AUC for quantitative studies with sample size ≥100 were 0.771 ( Figure 6. Forest plots of SEN (a) and SPE (b) for diagnostic value of ctDNA assay for CRC in the SEPT-9 methylation detection subgroup. 95% CI, 95% confidence interval; CRC, colorectal cancer; ctDNA, circulating tumor DNA; SEN, sensitivity; SPE, specificity; diamond shapes representing average effect; dark blue squares representing individual study results.

Subgroup Analysis and Meta-Regression Analysis
Different covariates including region (Asia vs. non-Asia), sample size (≥100 vs. <100), control type (health control vs. non-CRC disease), sample source (plasma vs. serum), assay methods (RT-PCR vs. other methods in quantitative analysis; MSP vs. other methods in qualitative analysis), and methylation gene (SEPT9 vs. other genes in qualitative studies) were applied in the subgroup analysis (Table 2).
Subgroup analysis based on sample size revealed an improved diagnostic accuracy in quantitative studies with sample size ≥ 100 since the SEP, SEN, and AUC for quantitative studies with sample size ≥100 were 0.771 ( For the quantitative analysis of ctDNA, subgroup analysis associated with control type showed that the SEN, SPE, and AUC for quantitative analysis of ctDNA to discriminate CRC from healthy control were 0.723 (95% CI: 0.626-0.803), 0.927 (95% CI: 0.833-0.967) and 0.861, while the corresponding values for quantitative analysis of ctDNA to discriminate CRC form non-CRC disease were 0.647 (95% CI: 0.508-0.765), 0.878 (95% CI: 0.563-0.976), and 0.641, respectively. While subgroup analysis in qualitative studies depicted an SEN of 0.601 (95% CI: 0.554-0.647), an SPE of 0.915 (95% CI: 0.895-0.932), and an AUC of 0.844 in qualitative studies with healthy people as control, and an SEN of 0.672 (95% CI: 0.617-0.723), an SPE of 0.809 (95% CI: 0.737-0.865) and an AUC of 0.774 in qualitative studies used non-CRC disease as control. These results indicated that studies utilizing healthy people as control provided a more satisfactory diagnostic efficiency compared to studies that used non-CRC patients as a control in both quantitative and qualitative analysis of ctDNA, which highlighted a more robust capability of ctDNA assay to distinguish CRC from healthy individuals than from non-CRC disease patients. Multiple levels of meta-regression also revealed that the parameter "control type" could contribute to the heterogeneity (Table 3). Moreover, heterogeneity was also derived from the parameter "methylation gene" as the corresponding coefficient was 0.661 (p = 0.01), which was consistent with the primary outcome of a clinical trial of SEPT9 methylation for CRC detection, indicating SEPT9 as a promising CRC diagnostic biomarker.

Publication Bias
Deek's funnel plot asymmetry test [39] was used to check the potential publication bias, and a p-value less than 0.05 indicated significance. No significant publication bias was observed in the quantitative analysis of ctDNA (Figure 7a, p = 0.307). Nevertheless, our results revealed a significant publication bias in the qualitative analysis of ctDNA (Figure 7b, p = 0.000). In addition, among these qualitative analysis studies, our analysis demonstrated that no significant publication bias existed in the SEPT9 methylation group (Figure 7c, p = 0.731), while a significant publication bias was found in the non-SEPT9 methylation group (Figure 7d, p = 0.000).

Publication Bias
Deek's funnel plot asymmetry test [39] was used to check the potential publication bias, and a p-value less than 0.05 indicated significance. No significant publication bias was observed in the quantitative analysis of ctDNA (Figure 7a, p = 0.307). Nevertheless, our results revealed a significant publication bias in the qualitative analysis of ctDNA (Figure 7b, p = 0.000). In addition, among these qualitative analysis studies, our analysis demonstrated that no significant publication bias existed in the SEPT9 methylation group (Figure 7c, p = 0.731), while a significant publication bias was found in the non-SEPT9 methylation group (Figure 7d, p = 0.000).

Discussion
CRC is one of the most common malignancies with poor prognosis and high mortality, which is highly correlated with the low early diagnostic rate [1]. When detected at an early stage, the survival rate of CRC would be over 90%, compared to approximately 7% for late-stage disease [120]. Although colonoscopy has demonstrated perfect diagnostic efficiency, it has low acceptance by general people due to the invasive procedure, troublesome diet restriction, and bowel preparation [10,11]. Thus, novel noninvasive diagnostic methods with robust diagnostic efficiency are in urgent need. Advance in precise ctDNA detection has made it possible for "liquid biopsy" of ctDNA to become a potential diagnostic biomarker of CRC [121]. In this diagnostic meta-analysis, we aimed to incorporate the diagnostic accuracy of ctDNA for CRC.
In our meta-analysis, quantitative analysis of ctDNA exhibited a higher SEN (0.723 vs. 0.609) and a comparable SPE (0.920 vs. 0.891) compared to qualitative analysis, which is consistent with a higher AUC (0.860 vs. 0.828). This might be explained by the fact that some genetic loci selected for the test were not commonly expressed in CRC, which might generate a false negative in the test. Various septins have been associated with tumorigenesis, and SEPT9 hypermethylation has been detected in CRC [122]. Furthermore, the detection of SEPT9 methylation in cfDNA has been regarded as a potential biomarker for

Discussion
CRC is one of the most common malignancies with poor prognosis and high mortality, which is highly correlated with the low early diagnostic rate [1]. When detected at an early stage, the survival rate of CRC would be over 90%, compared to approximately 7% for latestage disease [120]. Although colonoscopy has demonstrated perfect diagnostic efficiency, it has low acceptance by general people due to the invasive procedure, troublesome diet restriction, and bowel preparation [10,11]. Thus, novel noninvasive diagnostic methods with robust diagnostic efficiency are in urgent need. Advance in precise ctDNA detection has made it possible for "liquid biopsy" of ctDNA to become a potential diagnostic biomarker of CRC [121]. In this diagnostic meta-analysis, we aimed to incorporate the diagnostic accuracy of ctDNA for CRC.
In our meta-analysis, quantitative analysis of ctDNA exhibited a higher SEN (0.723 vs. 0.609) and a comparable SPE (0.920 vs. 0.891) compared to qualitative analysis, which is consistent with a higher AUC (0.860 vs. 0.828). This might be explained by the fact that some genetic loci selected for the test were not commonly expressed in CRC, which might generate a false negative in the test. Various septins have been associated with tumorigenesis, and SEPT9 hypermethylation has been detected in CRC [122]. Furthermore, the detection of SEPT9 methylation in cfDNA has been regarded as a potential biomarker for CRC diagnosis and has become the first blood-based CRC screening test approved by the FDA [123]. When we looked into the SEPT9 methylation analysis among the qualitative analysis, the corresponding AUC was 0.883, relatively higher than the AUC of the quantitative analysis, with the value of the SEN (0.679) and SPE (0.904) similar compared to the quantitative analysis. These results depicted the superiority of SEPT9 methylation detection over other gene locus analyses in CRC diagnosis.
DOR was also applied to evaluate the diagnostic accuracy, and the discriminatory test performance would be classified as satisfactory when the value of DOR was over 10 [31]. Both quantitative and qualitative analyses of ctDNA exhibited satisfactory discriminatory test performance as the value of DOR was 23.305 and 12.621, respectively. Specifically, SEPT9 methylation assay among qualitative analysis revealed a higher DOR of 20.551 than that of qualitative analysis of ctDNA, indicating a better discriminatory test performance of SEPT9 methylation assay than other single gene methylation assay in qualitative analysis.
In this study, the value of PLR of quantitative analysis and SEPT9 methylation were 6.820 and 6.420, respectively, indicating that CRC subjects have an approximately six to seven folds higher chance of being positive in quantitative analysis and SEPT9 methylation analysis of ctDNA. However, qualitative analysis of ctDNA depicted a PLR of 4.920, which is lower than the value of quantitative analysis and SEPT9 methylation. Moreover, the qualitative analysis of ctDNA yielded an NLR of 0.496, implying that the probability for cases with negative qualitative assay results to have CRC is 49.6%, whereas the value of NLR of quantitative analysis and SEPT9 methylation detection of ctDNA were 0.326 and 0.367, respectively. These results might be explained that some gene locus methylation might not exist in all CRC cases, and using single gene methylation as a biomarker might lead to false negatives. However, SEPT9 methylation demonstrated a comparable PLR and NLR with quantitative analysis of ctDNA, depicting the robust diagnostic performance of SEPT9 methylation assay. Additionally, the SEPT9 assay demonstrated better compliance than FIT and colonoscopy in CRC screening [8,9].
Publication bias was evaluated by Deek's funnel plot asymmetry test, and no publication bias was revealed in the quantitative analysis of ctDNA (p = 0.307). However, it was highly possible that publication bias was presented in the qualitative analysis group (p = 0.000). As previous analysis demonstrated the disparity between the SEPT9 methylation assay and other gene methylation assays within the qualitative analysis, we further evaluated the publication bias based on gene loci. The Deek's funnel plot asymmetry test revealed that publication biases were likely to exist in the non-SEPT9 methylation assay (p = 0.000) but not in the SEPT9 methylation assay (p = 0.731). Furthermore, a metaregression analysis was adapted to explore the potential source of heterogeneity. In the quantitative analysis of ctDNA, none of the parameters of region, control types, sample size, ample source, and assay methods represented a primary source of heterogeneity; thus, heterogeneity might be generated from subjects' ages, TMN stages, tumor size, tumor metastasis, or different protocol, which failed to be evaluated in this meta-analysis due to the lack of complete information. In the qualitative analysis of ctDNA, among parameters of region, control types, sample size, ample source, assay methods, and gene loci methylation, "control type" (coefficient = 0.502, p = 0.034) and "gene loci methylation" (coefficient = 0.661, p = 0.010) were potential primary sources of heterogeneity. We further analyzed the heterogeneity in SEPT9 methylation studies and revealed that "region" (coefficient = 0.676, p = 0.045) and "control type" (coefficient = 0.733, p = 0.035) could be the primary sources of heterogeneity. Thus, further large clinical trials should reasonably select the control subjects and various ethnicities or be performed in multiple regions to boost the diagnostic performance of ctDNA of CRC.
Remarkably, there were several limitations that should be discussed in this meta-analysis. Firstly, even though we searched various databases, there were several publications we did not include in our meta-analysis as we failed to access the full texts. Moreover, there were only a small number of studies that described quantitative analysis of ctDNA, which might diminish the statistical significance. Thirdly, we failed to include some parameters, such as TMN stages, as incomplete information in included articles. Additionally, potential publication bias within the qualitative analysis that detects non-SEPT9 methylation, which might diminish the confidence of the analysis. Eventually, single gene loci were analyzed in qualitative analysis, while the combination of different gene alterations or other detection techniques might help to improve the diagnostic performance [124]. Therefore, more large-scale clinical trials with complete information should be performed to delineate the diagnostic value of ctDNA detection for CRC to further clarify the conclusions in this meta-analysis.

Conclusions
In summary, this is the first integrated meta-analysis on the overall diagnostic accuracy of circulating tumor DNA assays in CRC. Both quantitative and qualitative analyses of ctDNA exhibited excellent diagnostic efficiency. Specifically, quantitative analysis of ctDNA or qualitative analysis of SEPT9 methylation could potentially serve as effective diagnostic methods for CRC. Larger sample studies are needed to further confirm our conclusions and to make the ctDNA approach more sensitive and specific.
Supplementary Materials: The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/jcm12020408/s1, Table S1: Summary of most relevant features of the enrolled publications. Figure S1: Traffic-light plot of quality assessment by the revised QUADAS-2. QUADAS-2, Quality Assessment of Diagnostic Accuracy Studies-2.