Diagnostic Accuracy of Methylated SEPT9 for Blood-based Colorectal Cancer Detection: A Systematic Review and Meta-Analysis

Colorectal cancer (CRC) is one of the most common malignant tumors and places an enormous burden on the society. It was estimated that 1.4 million new cases were diagnosed worldwide in 2012,1 of which, more developed countries accounted for the larger proportion. In contrast to incidence, mortality rates of CRC have been found to decrease in numerous countries, which most likely benefits from early detection.2 It is predicted that a total of 277,000 new CRC cases and 203,000 CRC-induced deaths in United States will be averted from 2013 to 2018 if National Colorectal Cancer Roundtable reaches the goal of increasing the prevalence of CRC screening to 80% by 2018.3 Although there are various guideline-recommended methods one can choose for detection, the compliance remains low. The data in 2013 showed that only about 57% of eligible adults adhered to screening recommendations provided by US Preventive Services Task Force (USPSTF).4 There are many reasons for low adoption for CRC detection. Obstacles specific to colonoscopy include aversion to bowel preparation, discomfort during the procedure, pre- and post-procedure time requirements, and costs.5 Guiac-based fecal occult blood tests or fecal immunochemical tests (FITs) are easier to be accepted. However, both methods continue to be underutilized and have relatively low diagnosis value.6 Since the currently utilized methods have various limitations and there is no other information available for detection, it is very important to introduce better and more patient-friendly approaches, especially blood testing, for detecting CRC.7 It is known that CRC occurs due to the genetic and epigenetic alterations of intestinal epidermal cells.8 Therefore, the determination of specific molecular markers targeting the changes may be a promising method for detecting early CRC. Aberrant methylation of tumor DNA sequences has been found in various genes, of which, methylated Septin 9 (mSEPT9) DNA is validated to be able to effectively diagnose CRCs from normal blood using real-time PCR.9 SEPT9, a member of the Septin family, has been found to function in cytokinesis and remodeling cytoskeletal.10 mSEPT9 was found to be correlated with carcinogenesis.10 Multiple research assays have been developed to identify mSEPT9 in circulating plasma by PCR amplification. A number of case–control studies, which encompassed thousands of clinical samples,9, 11, 12, 13 have been performed to verify the accuracy of mSEPT9 for CRC detection. In these studies, the sensitivity and specificity ranged from 69 to 79% and 82 to 99%, respectively. However, a prospective study (PRESPET NCT00855348) published later in 2014, which recruited almost 8000 samples, showed that the sensitivity was only 50.9%, lower than the expected data.14 Until then, it still lacked convincing evidence to translate such methods from research into clinical practice. Given that determination of mSEPT9 in blood has a promising future for CRC screening, existing researches and guidelines still fall short of giving detailed instructions to improve clinical applications which may be a result of insufficient evidence or underestimated diagnostic value. There are various methods (MethyLight, MSP-DHPLC, MS-HRM) and assays used in detecting mSEPT9, most of which are claimed to have high value. Epi proColon itself has two generations of assays and three inspection methods. The limitations above may hinder the understanding of optimal utilization strategy until more accurate and detailed explanations are provided. Therefore, we have performed a systematic review and meta-analysis of the diagnostic accuracy of mSEPT9 in order to explore the optimal method and kit for CRC detection. To estimate mSEPT9 in peripheral blood, the index test should be the methods and kits used, while the reference test should be colonoscopy. Any studies that estimated mSEPT9 in stools or other tissues were not included, neither were the ones using other comparator tests. We manually retrieved and examined the reference lists of relevant articles for additionally eligible studies. We also searched OpenGrey.eu for potential grey studies and clinical trials registry platforms such as ICRTP for ongoing and recently completed ones. If significant heterogeneity existed, we investigated the causes of heterogeneity by performing subgroup analysis and meta-regression when sufficient studies were available. The following categorical covariates were used: assays or methods of index test, race, CRC stage and differentiation, pathology types, etc. Spearman correlation coefficients between sensitivity and 1-specificity were also estimated for the threshold effect. Furthermore, Deeks’ funnel plot was used to estimate the risk of publication bias, and a P value <0.05 indicated high risk of bias. Eight studies showed high concern of applicability for the reason that they only enrolled healthy persons in control group. Seven studies had unclear concern because the threshold and assay were not interpreted in details. All of the studies showed low concern about reference standard. Furthermore, subgroup analysis was therefore performed by ethnicity, study type, assay, tumor stage and differentiation, combined method and precancerosis (see Table 2). We also conducted subgroup analysis based on assay or method which was used in included studies and the results equaled that of Epipro Colon assay and other methods (MethyLight, MSP-DHPLC, MS-HRM, etc.) but differed between generation 1 and generation 2 Epipro colon assay. The pooled sensitivity was 0.76 and the specificity was 0.94 in the generation 2 assay, higher than that of generation 1. In addition, data was further extracted and analyzed by the groups of disease stages and combined methods. The pooled sensitivity, specificity, LR+, LR−, DOR, and AUC are 0.79, 0.93, 11.0, 0.22, 49, and 0.92 in stage IV, respectively, which shows the highest diagnosis value, followed by stages III, II, and I. Similarly, CRC cases with low differentiation were more likely detected than moderate and high one. Three studies combined mSPET9 with FIT in parallel tests to estimate diagnosis accuracy and the results showed higher sensitivity (0.94) and lower specificity (0.68) than using mSPET9 alone. There was not enough data to combine carcinoembryonic antigen (CEA) or other methods in testing diagnostic accuracy. Twelve studies provided the details about results in adenomas and polyps. The pooled sensitivity was 0.15 and 0.05 in adenomas and polyps, respectively, both indicating low positive ratio of mSPET9 detection. Moreover, the pooled sensitivity was 0.23 for larger size (large than 1 cm) polyps or adenomas, which is higher than smaller ones (0.09; see Table 2). Since Figure 3 indicates significant heterogenity of sensetivity and specificity after computing the Cochran’ Q statistic and I2 (both P value <0.05), meta-regression was therefore conducted to trace the causes. The result shows that study types, kits used (Epipro colon or not), country (Asia or not), sample size (> or <300) and risk of bias of included studies all lead to the heterogeneity of sensitivity and specificity in a single variable model, of which whether the studies were performed in Asian countries or not was significant in joint model (P=0.01; see Figure 5). Figure 6 presents symmetry in Deeks’ funnel plot (P=0.41) and indicates that there exists no significant publication bias in the included studies. Recently, USPSTF updated its recommendations and initially reviewed the evidence on the efficacy of detection CRCs with mSEPT9.36 In our systematic review, we estimated that the pooled sensitivity and specificity was 0.71 and 0.92, respectively, proving to be reliable for CRC detection. The results were apparently higher than those in PRESEPT study,14 which may owed to recruiting early asymptomatic CRC patients for analysis. The systematic review also performed stage and differentiation-related analysis in detection, and Table 2 presents an apparent positive correlation between the detection rates of CRC and stage degrees. The results indicates that advanced stage CRCs are easier to be detected by mSEPT9 than early stage. The trend was similarly observed in tumor differentiation. Low-differentiation CRCs has much higher sensitivity than high differentiation ones. The results showed Asia Group had higher sensitivity than other continents. However, the results from Korea18 showed obvious lower sensitivity (0.363). The discrepancy might have occurred due to the potential racial differences and kit variations.37 In our subgroup analysis, we tried to explore the optimal method and assay for mSEPT9. 20 studies investigated the accuracy of Epi proColon and only four of included studies focused on other assay kits (mainly using the MytheLight method). Both assays presented similar results, but the Epipro Colon was found to be described in details and thus easier for clinicians to operate. The second generation of Epipro Colon has received approval from the US Food and Drug Administration38 and was reported to have resolved many technical hurdles and improved in several aspects, such as employing a novel bisulfite DNA conversion and purification technology39 as well as a new real-time PCR reaction.13 Two different types of algorithms were applied for Epicolon Colon in the studies and the results were different in sensitivity and specificity. Sensitivity was high using a 1/3 algorithm test but the specificity was low. Although sensitivity was low using a 2/3 algorithm test, it had a high true negative rate. Since it is more important to improve the capability in excluding non-cancer samples and avoiding the rate of misdiagnosis, 2/3 algorithm is recommended for CRCs detection. As a first blood-based detection method recommended for CRC, can mSEPT9 really improve compliance? The data results from a German research ensured the practicability, in which 83% of patients were willing to accept mSEPT9 test, which is higher than colonoscopy (37%) and stool test (15%).40 Even though the systematic review concluded an encouraging result of mSPET9 in CRC detection, it still has several limitations. First of all, FIT is currently widely used in CRC screening. However, due to lack of appropriate studies for further analysis, we did not provide further information about sensitivity and specificity in comparison between mSPET9 and FIT. Secondly, despite the diagnostic value of detecting advanced stage CRCs (III–IV), the analysis that were focused on early stage of CRC (Stage I) and adenomas or polyps showed low sensitivity. It turned out the diagnostic value of mSEPT9 may, to some degree, be limited in precancerous lesions and CRC in Stage I. However, mSEPT9 was shown to have low misdiagnosis rate and sensitivity may be improved when combined with FIT. Thirdly, as different methods were used for detecting mSEPT9, we did not subgroup analyze the optimal threshold for every method other than Epipro Colon. Three different cut-off points were used for this assay, of which Ct<45.0 was the most utilized. The sensitivity was 0.70 when Ct<45.0 was used, slightly lower than Ct<41.0, indicating Ct<45.0 may be more sensitive for utilization. But it still need further study to verify it as the best threshold. Fourthly, this meta-analysis did not include any language other than Chinese and English. Restriction in languages may bring about a potential risk of publication bias. In terms of methodological quality, most studies that were included were case–control in design and excluded “difficult-to-diagnose” patients, which may lead to a risk of bias in patient selection and overestimation of diagnostic accuracy.18 Finally, although it was reported that mSEPT9 could be employed as a predictor of CRC recurrence, metastasis and survival,18, 41 there is insufficient data for synthesis in our meta-analysis in order to draw robust conclusions about the value as a follow-up marker. In conclusion, our systematic review suggests that mSEPT9 can be used as an effective marker for blood-based CRC detection. Based on current evidence, the second generation Epipro Colon (Epigenomics) could be used as the optimal assay kit with 2/3 algorithm. In addition, the review revealed that a larger sample size and more prospective studies were needed to further verify the diagnostic value of mSEPT9. Guarantors of the article: Mingwei Yu, MD; Ganlin Zhang, MD; and Xiaomin Wang, PhD. Specific author contributions: Jiayun Nian, MD and Xu Sun, MD contributed to the study design, data extraction and interpretation, and drafting and final approval of the manuscript. Su Yang Ming, MD and Chen Yan, MD contributed to selection of studies and final approval of the manuscript. Yunfei Ma, MD and Ying Feng, MD contributed to data extraction and final approval of the manuscript. Lin Yang, MD contributed to study appraisal and final approval of the manuscript. Financial support: The Beijing Municipal Science and Technology Plan Projects (NO. D161100005116005); Natural Science Foundation of China (NO. 81473643). Potential competing interests: None. Trial Register Number: CRD42016042457. Supplementary Information accompanies this paper on the Clinical and Translational Gastroenterology website