Diagnostic value of RASSF1A methylation for breast cancer: a meta-analysis

Background: Numerous studies reported that RAS-association domain family 1 isoform A (RASSF1A) methylation might act as diagnostic biomarker for breast cancer (BC), this meta-analysis aimed to evaluate the value of RASSF1A methylation for diagnosing BC. Methods: Such databases as PubMed, Cochrane Library and Web of Science databases were searched for literatures until May 2019. A meta-analysis was performed utilizing STATA and Revman softwares. Furthermore, subgroup analysis was adopted to determine likely sources of heterogeneity. Results: Totally 19 literatures with 1849 patients and 1542 controls were included in the present study. Sensitivity, specificity, diagnostic odds ratio (DOR) and the area under the summary receiver operating characteristic curve (AUC) of RASSF1A methylation for diagnosing BC were 0.49, 0.95, 19.0 and 0.83, respectively. The sensitivity (0.54 vs 0.43), DOR (30.0 vs 10.0) and AUC (0.84 vs 0.81) of RASSF1A methylation in Caucasian were higher than other ethnicities. The sensitivity (0.64 vs 0.57), DOR (21.0 vs 14.0) and AUC (0.89 vs 0.86) of methylation-specific PCR (MSP) were superior to other methods (q-MSP, OS-MSP and MethyLight). The sensitivity, DOR and AUC of serum RASSF1A methylation vs RASSF1A methylation in other samples (tissue or plasma) were 0.55 vs 0.40, 22.0 vs 14.0 and 0.86 vs 0.74, respectively. Conclusions: RASSF1A methylation might be a potential diagnostic biomarker for BC. Considering its low sensitivity and high specificity, it should combine with others to upgrade the sensitivity. Besides, under such conditions, MSP detection, serum RASSF1A methylation and Caucasian are shown to be more effective and suitable for diagnosing BC.


Introduction
Breast cancer (BC) is the most popular cancer and the most common reason for cancer death in female worldwide [1]. The development of breast cancer is a multi-step procedure, and its prevention is still a difficult task across the world. It is well known that early diagnosis is beneficial, in some advanced countries, the 5-year relative survival rate for BC patients is beyond 80% as a result of early diagnosis. Thus, early screening of BC among high-risk subjects represents a significant way to upgrade prognosis of the patients [2]. It is general practice among primary care, gynecology and oncology physicians as well as surgeons to carry out clinical examination when BC is suspected. Recent research reports numerous diagnostic biomarkers, but there is no clinical guideline or unanimous consensus with regards to the use of biomarkers for early diagnosis of BC [3]. Therefore, effective biomarkers are truly needed.
Many studies currently demonstrate that the detection of methylated circulating cell-free DNA in patient's peripheral blood might be a favorable quantitative and non-invasive method for diagnosing BC [4]. The promoter region methylation of tumor suppressor genes causes transcriptional silencing. Many significant genes that suffer from transcriptional silencing are involved in critical cancer-related cellular pathways. Studies demonstrate that the patterns of gene methylation are distinctive and the abnormal methylation of genes could act as diagnostic biomarkers for BC [5].
RAS-association domain family 1 isoform A (RASSF1A), belonging to family of RAS effectors, is a tumor suppressor gene coding particular protein. RASSF1A is expressed in normal breast cells while inactivated in breast tumors due to DNA methylation of their CpG island in promoters, RASSF1A displays high promoter methylation of 56% in BC but 8% in normal tissues. The result suggests that DNA methylation of RASSF1A may be common in BC [6,7].
Increasing evidence shown the relationship between RASSF1A and BC, but the value of RASSF1A methylation in the diagnosis of BC is uncertain [8]. Mohammad et al. reported that the sensitivity of RASSF1A methylation was 32%, but Noriaki et al. showed the sensitivity was 90% [9,10]. The heterogeneity was high among different studies . In the present study, we used a meta-analysis to identify the diagnostic value of RASSF1A methylation for BC. Two independent reviewers screened the literatures in line with the inclusion and exclusion criteria. The inclusion criteria included: (1) The research must investigate the diagnosis value of RASSF1A methylation for BC. (2) The patients were diagnosed with BC by pathology. (3) Literatures were published in English. (4) Sensitivity and specificity of RASSF1A were offered abundantly to build 2 × 2 contingent tables. The exclusion criteria were made as follows:

Search strategy and study selection
(1) Studies were categorized as reviews, letters, guidelines, consensus statements, meeting abstracts and words of editors. (2) Studies lacked enough data, such as the RASSF1A methylation could not be extracted or computed from the original research.

Data extraction and quality assessment
Two reviewers extracted following data from the eligible literatures: first author, year of publication, sample size, sample type, detection method and ethnicity. When disagreement was presented, a consensus was realized after discussion with a third reviewer. Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) was utilized to evaluate the quality of studies [11]. QUADAS-2 represented an evidence-oriented quality assessment approach including four areas: patient selection, index test, reference standard, and flow and timing.

Trial Sequential Analysis (TSA)
Trial Sequential Analysis (TSA) was performed by the TSA software (Centre for Clinical Intervention Research, Copenhagen Trial Unit). Two-sided type I error of 5% and two-sided type II error of 5% were used in conventional test and α-spending boundaries. 'Information size' , 'Relative risk reduction' and 'Incidence in Intervention Group' were automatically generated by the TSA software, and 'Incidence in Control Group' = 5% were set to calculate the required sample size. If the cumulative Z-curve crossed the monitoring boundaries constructed by both information size calculations and reached the required sample size line, the number of samples included in the meta-analysis is sufficient and the result is reliable.

Statistical analysis
First, we measured the threshold effect and heterogeneity. If the Spearman correlation coefficient's P-value was over 0.05, there was no threshold effect, then a heterogeneity produced through non-threshold effect possessed a further analysis. If P < 0.05 or I 2 > 50%, the heterogeneity was significant and a random-effect model was built. Subgroup analysis were performed to find the sources of heterogeneity. The four covariates for subgroup analysis were as follows: (1) ethnicity of patients, (2) sample size, (3) the detection method and (4) the sample type. When P exceeded 0.05 and I 2 was not higher than 50% in one subgroup, the related covariate was seen as the source of heterogeneity. The pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR) and diagnostic odds ratio (DOR), along with their 95% confidence intervals (Cls) were calculated and graphically shown in forest plots. A summary receiver operating characteristic (SROC) curve was built and the area under the SROC curve (AUC) was

Search results
A total of 893 literatures were searched from PubMed, Cochrane Library and Web of Science databases. No new literature was observed after looking for the references. Altogether, 874 literatures were excluded as they did not satisfy the inclusion criteria: 269 literatures were excluded owing to duplication, 443 literatures were excluded after screening the title and abstract; 162 literatures were excluded after assessing the full-text, Finally,19 literatures were included in this meta-analysis ( Figure 1). Three literatures analyzed two different studies and eventually 22 studies were included in the meta-analysis.

Publication bias and quality assessment
No asymmetry was found in the Deek's funnel plot and P > 0.05 indicated no publication bias (Supplementary Figure  S1). The outcome of the QUADAS-2 (Supplementary Figure S2) showed high risk of bias on patient selection while low or unclear bias on index test, reference standard and flow and timing. So, the entire qualities of include studies were medium.

TSA
The cumulative Z-curve (Supplementary Figure S3) crossed the monitoring boundaries and the number of samples (n = 3391) has exceed the required sample size (n = 411), confirming that our sample sizes were enough and the estimates are reliable.

Heterogeneity and subgroup analysis
In the threshold analysis, spearman correlation coefficient was -0.053 and P-value was 0.814, showing no threshold effect. Furthermore, a significant heterogeneity caused by non-threshold effect was seen. The result indicated that I 2 of sensitivity was 96.33%, I 2 of specificity was 96.76% and a random-effects model was utilized. To determine the sources of heterogeneity, we conducted subgroup analysis.  Table 2).

Discussion
As an important tumor suppressor gene, RASSF1A has been reported as a new diagnostic biomarker for BC in numerous studies [8,9,12,13,32]. However, no comprehensive review had been performed to evaluate the diagnostic accuracy. This is the first diagnostic meta-analysis to assess the diagnostic value of RASSF1A methylation for BC. The pooled sensitivity and specificity of RASSF1A methylation in our study was 0.49 and 0.95, respectively, which indicated that 49% of BC patients enjoyed high RASSF1A methylation levels, and 95% of non-breast-cancer patients   had low RASSF1A methylation levels. We identified that RASSF1A methylation was a potential diagnostic biomarker for BC; but the sensitivity was low, thus it should be combined with other biomarkers to increase sensitivity. Additionally, the SROC curve and AUC can be made to assess diagnostic value, where AUC is closer to 1 signifies that the diagnostic method has better discrimination [33]. The AUC performed in the current meta-analysis was 0.83, showing that RASSF1A methylation is a helpful biomarker for BC diagnosis. The DOR was another indicator of diagnostic accuracy that comprehensively indicating sensitivity and specificity. The higher the value of DOR, the better the diagnosis effort was [34]. A DOR of 19.0 in our study displayed a favorable value of RASSF1A methylation. The PLR and NLR were other diagnostic indicators of clinical importance, PLR close to 10 and NLR close to 0 were thought more convincing to rule in or rule out disease, respectively [35]. In our research, the pooled PLR and NLR were 10.03 and 0.53 that have medium value of RASSF1A methylation. Several DNA methylation genes were reported in BC; however, the most suitable diagnostic marker hasn't been found yet [4,36]. The specificity of RASSF1A methylation was higher than others. But its sensitivity was lower. For example, Ming et al. showed that the specificity of RASSF1A methylation was 91.95%, while the specificity of P16 methylation, PCDHGB7 methylation, SFN methylation and HMLH1 methylation were 84.32%, 54.66%, 38.98% and 77.97%, respectively [37]. The sensitivity of RASSF1A methylation was 17.16%, while the sensitivity of P16 methylation, PCDHGB7 methylation, SFN methylation and HMLH1 methylation were 22.39%, 55.60%, 73.51% and 27.99%, respectively [9]. Given its low sensitivity, a combination with other gene methylation may upgrade the sensitivity and increase its utility in diagnostic test. Recently, Mohammad et al. showed that RASSF1A methylation combing GSTP1 methylation and RARβ2 methylation improved sensitivity for the diagnosis of BC (sensitivity: 62%, specificity: 87%). Kim et al. reported that the combination with Twist methylation, RARβ methylation and RASSF1A methylation could also improve sensitivity for the diagnosis of BC (sensitivity: 96%, specificity: 81%) [10,[14][15][16][17]24]. The subgroup analysis was performed according to the ethnicity of BC patients, sample size, detection method and sample type. The pooled sensitivity, DOR and AUC of RASSF1A methylation diagnosing in Caucasian were 0.54, 30.0 and 0.84, while the pooled sensitivity, DOR and AUC of RASSF1A methylation in other ethnicities were 0.43, 10.0 and 0.81, respectively. The results indicated that Caucasian might be more susceptible to RASSF1A methylation. It suggested that genetic and environmental factors had impact on the diagnostic value of RASSF1A methylation [37]. When the sample size is expanded from <100 to ≥100, the pooled sensitivity (0.40 vs 0.51), DOR (12.0 vs 21.0) and AUC (0.72 vs 0.87) of RASSF1A methylation increased, suggesting that higher diagnostic value of RASSF1A methylation was generated after expanding the sample size. The pooled sensitivity, DOR and AUC of RASSF1A methylation detected by MSP were 0.64, 21.0 and 0.89, while the pooled sensitivity DOR and AUC of RASSF1A methylation detected by other methods were 0.57, 14.0 and 0.86, respectively. The results indicated that MSP was more effective in the detection of RASSF1A methylation. The reason of it might be that MSP and other methods required particular gene sequence information for the design of PCR primers. The different primers might have impacts on results [38]. The pooled sensitivity, DOR and AUC of serum RASSF1A methylation were 0.55, 22.0 and 0.86, while the pooled sensitivity, DOR and AUC of RASSF1A methylation in other sample types were 0.40, 14.0 and 0.74, respectively. The outcomes indicated that the detection of serum RASSF1A methylation was more suitable for diagnosing BC compared with tissue or plasma RASSF1A methylation.
Our meta-analysis has several disadvantages. First, all studies here were case-control studies, which have higher risks of bias. So more in-depth investigations with well-designed prospective cohort studies are needed. Second, some patients with multiple genes methylation included in the meta-analysis might have an impact on the generalizability of the findings. Third, we only reported one gene RASSF1A methylation in the present research and more candidate gene methylations should be excavated in the future. Fourth, the clinical heterogeneity and statistical heterogeneity might decrease the reliability of our outcomes. Fifth, we can't provide our own experimental data for further verification. The following studies using both data from TCGA or laboratories are recommended to validate the result. This is the first diagnostic meta-analysis investigating the accuracy of RASSF1A methylation for BC. Our evidence shows that RASSF1A methylation enjoys narrow applicability for diagnosing BC yet. To improve diagnostic accuracy of RASSF1A methylation, it should be combined with other biomarkers. The result may be influenced by ethnicity, detection method and sample type used. The detection of RASSF1A methylation in Caucasian is more susceptible, MSP was more effective in the detection of RASSF1A methylation and serum RASSF1A methylation was more suitable for diagnosing BC compared with tissue or plasma RASSF1A methylation. More well-designed prospective diagnostic studies are needed to confirm our results.