A literature search for eligible studies that explored the association between the pri-miR-124-1 rs531564 polymorphism and cancer risk was carried out using PubMed, EMBASE, Web of Science and CNKI databases (until October 1, 2014). Keywords used in the searches included: (‘miR-124’ or ‘miRNA-124’ or ‘rs531564’ or ‘pri-miR-124-1’) and (‘tumor’ or ‘cancer’ or ‘carcinoma’) and (‘polymorphism’ or ‘SNP’ or ‘allele’ or ‘variation’). The search was performed without restriction on language or publication years. The reference lists of retrieved studies were also checked for additional literature.

Studies were identified as eligible if they met the following criteria: i) Evaluated the association of the pri-miR-124-1 rs531564 polymorphism and cancer risks; ii) case-control studies; and iii) detailed genotype data for estimating odds ratios (ORs) and 95% confidence intervals (CIs). Studies were excluded based on the following criteria: i) Reviews or animal studies; ii) lack of sufficient data for meta-analysis; and iii) repeated outcomes from the same samples. When overlapping data of the same case series were included in more than one study, only the most recent or complete study was used in the meta-analysis.

The information extracted from each study was as follows: First author, publication year, country origin, ethnicity, cancer types, diagnostic methods for cancer, source of controls, genotyping methods, the numbers of genotyped cases and controls and Hardy-Weinberg equilibrium (HWE) for control groups. Quality assessment of the included studies was evaluated according to the Newcastle-Ottawa Quality Assessment Scale (NOS) (18). The information was extracted by two investigators (Fang and Zeng) in duplicate and any disagreements were discussed by group discussions to achieve a consensus.

The pooled ORs and 95% CIs were summarized to assess the strength of the association between the pri-miR-124-1 rs531564 polymorphism and cancer susceptibility with five genetic models: Allele contrasts (G vs. C), and homozygote (GG vs. CC), heterozygote (CG vs. CC), dominant (GG/CG vs. CC) and recessive models (GG vs. CG/CC), separately. Heterogeneity was examined with the χ² and I² test. When heterogeneity was absent (P>0.10, I²<50), a fixed effect model was used for secondary analysis. Otherwise, a random effect model was employed (19,20). HWE was measured with the Pearson's goodness-of-fit test in control groups and P<0.05 was deemed not to conform to HWE. Sensitivity analysis was conducted to assess the stability of the combined results by the omission of every single study each time or excluding studies with disrupted HWE. Funnel plots were also used to estimate the possible publication bias. All the analyses were operated using Cochrane Collaboration's Review Manager Software 5.2 and P<0.05 was considered to indicate a statistically significant difference.

Based on the selection criteria above, five case-control studies (13–17) were finally enrolled in the meta-analysis with publication years that ranged from 2011 to 2014. The detailed literature retrieval process is presented in Fig. 1. A total of 4,763 subjects were involved in the meta-analysis, including 2,253 cancer cases and 2,510 healthy controls. Multiple cancer types were analyzed in the studies, including triple-negative breast cancer (13), cervical cancer (15) and ESCC (14,16–17). The diagnoses of these tumors were confirmed by histological or histopathological examinations. In addition, three studies used hospital-based controls (13–15), one used population-based controls (17) and one used population- and hospital-based controls (16). Polymerase chain reaction-based ligase detection reaction was performed in two studies (14,15) and three used other methods. The genotype frequencies of the controls were all fitted in HWE, except for one study by You (16). The NOS scores of the selected studies were >5 (moderate-to-high quality). The main characteristics and methodological quality of the included studies are summarized in Table I.

In the present analysis, data from five case-control studies were collected and analyzed to investigate the association between the pri-miR-124-1 rs531564 polymorphism and cancer risk. As a result, the meta-analysis of overall studies revealed that a significantly decreased cancer risk was observed in the G vs. C, GG vs. CC and GG vs. CG/CC models tested (G vs. C: OR, 0.86; 95% CI, 0.77–0.96; GG vs. CC: OR, 0.52; 95% CI, 0.34–0.79; GG vs. CG/CC: OR, 0.54; 95% CI, 0.36–0.81; Fig. 2). In the subgroup analysis by cancer types, the results showed that the rs531564 genotype was markedly associated with a decreased risk of ESCC (G vs. C: OR, 0.87; 95% CI, 0.77–0.98; GG vs CC: OR, 0.54; 95% CI, 0.34–0.84; Fig. 3). The main results are summarized in Table II.

Moderate heterogeneity was observed in the CG vs. CC and GG/CG vs. CC models tested [CG vs. CC: P=0.22, P-value of Q test for heterogeneity test (Ph)=0.03, I²=62%; GG/CG vs. CC: P=0.10, Ph=0.08, I²=52%; Table II]. Therefore, the random effect models were used and sensitivity analyses were performed. When the study by Xiong et al (15) was removed, the heterogeneity in the GG/CG vs. CC model was largely reduced, but the result remained insignificant (P=0.13, Ph=0.25, I²=26%). When the study with disrupted HWE by You (16) was removed, the heterogeneity in the GG/CG vs. CC model was reduced and the result was consistently insignificant (P=0.15, Ph=0.12, I²=49%). Sensitivity analysis indicated that the pooled ORs were not altered by deleting one study at a time (data not shown).

Funnel plots were performed to access the publication bias of eligible studies, in which the standard error of log (OR) for each study was plotted against its log (OR). The graphical funnel plots (Fig. 4) appeared symmetrical and no clear evidence of publication bias was observed among all the genetic models.