Two investigators independently conducted a literature search using PubMed, Embase and Google scholar (last search updated on September 14, 2014). The keywords used included: BRCA1, breast carcinoma, breast cancer, methylation, prognosis and survival. In addition, the PubMed additional function: Related citations; and the references of the selected studies were scrutinized to identify additional studies.

Studies were included in the meta-analysis only if they had met the following criteria: i) Evaluated prognostic risk of patients with BRCA1 methylation; ii) provided overall survival (OS) or disease-free survival (DFS); iii) hazard ratios (HR) or odds ratios (OR) with its 95% confidence intervals (CIs); iv) published in English; and v) data from human subjects. In addition, studies were excluded if: i) Data was from reviews or animal studies; and ii) studies had the same population resources or overlapping datasets.

Following the exclusions, 9 studies met all the criteria. Two investigators independently extracted the following data from each study: First author's last name, year of publication, population, number of study subjects, effects on clinical outcomes (OS and DFS), and the number of methylated and unmethylated patients with a different status of ER, PR, HER2 and triple-negative receptors. OS is a term that denotes the chances of remaining alive for a group of individuals suffering from a type of cancer. At a basic level, the OS is representative of cure rates. DFS was defined as the chances of staying free of disease following a particular treatment for a group of individuals suffering from a type of cancer. It is an indication of how effective a particular treatment is.

Random effects and subgroup meta-analysis were performed according to the DerSimonian Laird method (13), due to the existence of heterogeneity between studies. The data were divided into two groups by population: European and Asian. The HRs were used to estimate the pooled effect of BRCA1 methylation on the prognosis of patients with breast cancer, and the ORs were pooled to estimate the strength of the association between BRCA1 hypermethylation and the risk of three negative statuses of receptors (ER, PR and HER2). When HRs (95% CIs) were shown only in the figure of survival curves, the authors were contacted for the exact value or the investigators estimated them according to the methods provided by Tierney et al (14). The Cochran Q (significant cut-off point: P=0.10) and I² (I²>50%, strong heterogeneity) statistics (15,16) were used to assess heterogeneity between studies. The Galbraith plot (17) was used to detect the potential sources of heterogeneity from the meta-analysis. Publication bias was assessed by funnel plot and the test of Egger et al (18). Sensitivity analyses were performed by the trim-and-fill method (19). All the analyses and graphs were obtained using STATA 11.0 software (StataCorp LP, College Station, TX, USA).

Fig. 1 summarizes the process of identifying eligible studies. Following the screening by two investigators independently, according to the inclusion criteria there were 9 studies with 3,131 study subjects entered into the meta-analysis (20–28). The characteristics of these studies are listed in Tables I and II. There were 4 studies from Europe and 5 studies from Asia.

As shown in Table II, there were 8 studies that met the inclusion criteria and were included in the present meta-analysis. The studies involved 337 BRCA1 promoter hypermethylations with an ER-negative status, 458 with a PR-negative status, 405 with an HER2-negative status, 169 with triple-negative receptors, and 780, 458, 1,140 and 258 controls without BRCA1 promoter methylation but with a negative status, correspondingly.

When the same investigators reported the results obtained from the same cohort of patients in several studies, only the largest series was included in the analysis. A cohort of patients was excluded due to duplicate studies.

Due to insufficient data, HRs on OS could be extracted from 7 studies for univariate analysis and 6 studies for multivariate analysis. According to DFS analysis, there were 4 studies with available data for univariate analysis and 5 studies with available data for multivariate analysis.

Considering the significant heterogeneity among studies (P=0.017, I²=61.1%), the random-effect model was used and a subgroup analysis was performed by considering different ethnicities or the population of the participants to estimate the combined effect of BRCA1 methylation. BRCA1 methylation was significantly associated with a poor OS and DFS of breast cancer in the univariate and multivariate analysis (Figs. 2 and 3 and Table III). The combined HR was 1.76 (1.15–2.68) and 1.97 (1.12–3.44) for univariate analysis and multivariate analysis of OS, respectively. For studies of DFS, the pooled HR was 1.28 (0.68–2.43) and 1.64 (0.64–4.19), respectively. The combined HRs (95% CIs) on OS by univariate analysis was 2.16 (1.21–3.86) for the Asian population, which was significantly higher than 1.21 (0.44–3.28) for the European population. All the combined HR scores in the present study are different from the study of Wu et al (9). They over-estimated the risk of fatality by using the OR instead of HR value from the Karray-Chouayekh study (29) and using the incorrect value of OS in the Xu et al study (23).

The majority of the studies identified that BRCA1 promoter methylation is correlated with poor survival, as shown in the present meta-analysis. However, an opposing opinion remains as BRCA1 promoter methylation is a protective factor in 2 studies [Krasteva et al (22) and Ignatov et al (21)]. To explore this difference, we identified that they used different primers. In the Krasteva et al (22) and Ignatov et al (21) studies, which found a better clinical prognosis in patients with BRCA1 promoter methylation, they used Baldwin's primer (30) while others used Esteller's primer (31). The two primer sequences were blasted to the human genome in the NCBI database to find the amplified regions. The two primers are for different regions where there was a different concentration of GC-rich regions. The Baldwin primer exhibited a larger amplification that contained more GC-rich regions and included the transcriptional start point (Fig. 4). Therefore, the difference of BRCA1 promoter methylation in the region of chr17: 43, 125, 429-43, 125, 541 (GRCh38 assembly) may affect the different prognosis. However, all of these require more testing by molecular-biological experiments to confirm.

The number of patients who carried BRCA1 methylation and one of the negative statuses of ER, PR, HER2 or triple-negative were extracted for the cases, as well as the number of patients without promoter methylation of BRCA1 correspondingly for the controls. The correlations between BRCA1 methylation and the negative status of different breast cancer-related receptors were meta-analyzed separately. The results of the association between hormone receptors that were negative, BRCA1 promoter methylation and the heterogeneity test are shown in Table IV. The overall results suggested that, particularly for the Asian populations, all the receptors negativity listed in Table II are associated with the BRCA1 promoter methylation. Patients with those estrogen, progesterone and epidermal growth factor-related negative receptors were more likely to be negative for the BRCA1 protein (OR>1.247, P<0.014). However, further molecular-biological experiments are required to determine whether the correlation is spurious, as the receptor and BRCA1 promoter methylation are associated with breast cancer.

The Patsopoulos method (32) was used to test if an individual study affected the heterogeneity in the OS and DFS analysis. As a result, if the Jing et al (28) and Sharma et al (26) studies were removed the heterogeneity disappeared (I²=17.3%, P=0.30). Funnel plot and Begg's test were performed to check the publication bias, and they suggest the absence of publication bias. The Begg's test for OS and DFS were P=0.187 and P=0.625, respectively.