No association between the progesterone receptor gene polymorphism (+331G/a) and the risk of breast cancer: an updated meta-analysis

We used four online electronic databases to select studies to include in this meta-analysis (PubMed, Web of Science, and Embase in English and China National Knowledge Infrastructure Database in Chinese; most recent search update, February 2017). Search terms included “breast cancer” or “breast neoplasm” or “mammary” combined with “progesterone receptor gene” or “PgR” or “+331G/A” or “+331G > A” or “rs10895068” and with “polymorphism” or “variant” or “genotype” or “allele”, without any limitation applied. Referenced lists of all included studies were then manually searched to identify any additional eligible studies. Only the study with the most recent, complete data was included when multiple studies included the same set of subjects.

Included studies met the following criteria: (1) case-control design; (2) clinical trial evaluating associations between +331G/A gene polymorphisms and breast cancer susceptibility; (3) pathological confirmation of breast cancer diagnosis was reported for all patients; (4) data regarding sample size and individual genotype frequencies were available for all cases and controls; and (5) at least two comparison groups (cancer group and control group) were included. Exclusion criteria: (1) duplication of prior studies and (2) meta-analysis, letters, reviews, or editorial articles.

Two investigators (Xing-ling Qi and Jun Yao) independently extracted data from eligible studies. Inconsistencies were resolved via discussion between the investigators. We recorded the first author’s name, publication year, country of origin, ethnicity studied, sample size, genotypes and allele frequencies for patients with the PgR +331G/A polymorphism, and Hardy-Weinberg equilibrium (HWE) results for controls groups. We recorded studies including more than one ethnicity as mixed ethnicity.

We used STATA 12.0 software (Stata Statistical software, College Station, TX, USA) to perform all statistical analysis. PRISMA checklists and guidelines were adhered to when performing the meta-analysis [18]. For control groups, we used Chi-square tests to analyze the Hardy-Weinberg equilibrium (HWE), with p < 0.05 indicating a significant deviation. Pooled frequency analyses were performed using Thakkinstian’s method [18, 19]. The strength of associations between the +331G/A polymorphism and breast cancer risk were evaluated using odds ratios (ORs) and 95% confidence intervals (CIs). Two-tailed tests were used to generate all p values.

We used five models to evaluate associations the +331G/A and breast cancer risk: allele model, dominant model, recessive model, homozygote comparison model, and heterozygote model. A random effects model was used to pool effect sizes of all included studies for a possible effect size across populations with different genetic backgrounds after considering the heterogeneity among the included studies [20]. We also used A as the risk allele to compare OR1 (AA vs. aa), OR2 (Aa vs. aa), and OR3 (AA vs. Aa) and further determined the genetic model that was the most appropriate under the instruction, as previously described [21, 22]. Heterogeneities among studies were estimated using an I² test, and describe I² values as low (25%), moderate (50%), or high (75%) estimates [22]. A Z-test resulting in a p value less than 0.05 determined statistical significance. We also explored the effect of included studies on combined ORs via sensitivity analysis employing sequential omission of each study. In addition, we conducted subgroup analyses by ethnicity (i.e. Caucasian, Asian, and mixed races) as well as by source of control subjects (i.e. hospital-based vs. population-based). We generated a funnel plot to reflect any possible publication bias [23, 24], with an Egger’s test resulting in ap < 0.05 indicating significant publication bias.

Table 3 shows our results generated using five genetic models to evaluate associations between the +331G > A polymorphism and breast cancer risk. Genetic model selection principles were used to determine the dominant model. Our summary results indicate that an association is indeed present between PgR +331G > A and the risk of breast cancer. Using a random effects model, we calculated a pooled OR of 1.140 (p = 0.027, 95% CI = 1.015–1.279) (Fig. 1).

We found no association between +331G/A polymorphism and breast cancer risk in Caucasian (p = 0.102, OR = 1.116, 95% CI = 0.978–1.272,), Asian (p = 0.105, OR = 1.749, 95% CI = 0.890–3.438) and mixed race (p = 0.231, OR = 1.170, 95% CI = 0.905–1.513) populations via subgroup analysis. Furthermore, using subgroup analysis by source of controls, there was an association between +331G/A locus and breast cancer risk in hospital-based (p = 0.004, OR = 1.295, 95% CI = 1.087–1.543,), but not in population-based controls (p = 0.440, OR = 1.046, 95% CI = 0.934–1.171; Table 4).

We examined the influence of individual studies the pooled ORs for +331G/A via sensitivity analysis involving omitting each study in each genetic model; the results did not change. This indicates that our results are statistically robust for all five genetic models examining associations between +331G/A and breast cancer susceptibility.

We assessed possible publication bias using a Begg’s funnel plot and Egger’s test. As shown in Fig. 2, no obvious asymmetry was observed in the funnel plot all genotypes in the overall population, and Begg’s test results did not reveal any publication bias (p > 0.05).