Characteristics of Included Studies
A total of 139 publications were preliminarily obtained by applying our searching strategy in five databases. Twenty-two studies were remained for further examination after the duplication removal and initial screening. While performing the final eligibility checking, we found that three ovarian cancer datasets from the Fourth People's Hospital of Hebei Medical University overlapped. According to our strategy, the two studies published earlier with smaller sample sizes were excluded, and the latest research data from December 2001 to December 2008 were retained.
Finally, 20 publications were qualified for the statistical analyses, including 4 in Chinese and 16 in English[19-39]. A total of 16,995 samples were enrolled, including 8,215 cancer patients in the case group and 8,480 healthy people in the control group. Among all the eligible papers, thirteen of them were conducted in China and seven others had people from Greece, Spain, Turkey, Sweden, Mexico, Polish and Brazil involved. And all studies were published between 2006 and 2019.The literature retrieval and screening process are shown in Figure 1. The basic characteristics of the included literatures are shown in Table 1.
Main Analysis Results
As demonstrated in Table 2, all the 95% confidence interval of the combined estimates contained OR = 1, indicating no statistically significant association between this polymorphism and cancer susceptibility (allele model: OR = 0.99, 95% CI 0.89-1.10; homozygous model: OR = 0.96, 95% CI 0.78-1.17; heterozygous model: OR = 1.00, 95% CI 0.91-1.10 Figure 2; dominant model: OR = 1.00, 95% CI 0.88-1.13; recessive model: OR = 0.98, 95% CI 0.83-1.16)
Subsequently, subgroup analyses for cancer type, origin area, genotyping method and sample size had been conducted to assess the impact of rs2252070 on cancer risk further. When stratified by cancer types, no significant association had been detected in the digestive system cancer group. Similarly, it revealed no significance in both lung cancer group and other cancer group. Next, in the analyses based on origin area and genotyping method, the pooled results also showed that this polymorphism had no significant effect on cancer incidence. Finally, we assessed the associations in sample size subgroups and set cut-off at 200, 400, 500, 600, 800 and 1000 respectively. However, the effect of rs2252070 polymorphism on cancer risk was still not statistically significant by applying all cut-offs.
Publication Analysis and Sensitivity Analysis
In order to validate the robustness of our study, the sensitivity analysis was conducted by leave-one-out method (Figure 3). By removing one certain study each time and re-calculating the summary statistic successively, the pooled ORs did not materially altered, showing that our results were not subject to any certain research.
In addition, when applying Begg’s test for appraisal of the publication bias, no significance in all genetic models had been detected. While when Egger’s test was used, there was significant publication bias in homozygous model (P = 0.030, Figure 4) and recessive model (P = 0.019, Figure 5). Asymmetric distribution was also observed in the corresponding funnel diagram. Using trim-fill method, four suppositional studies were added in homozygous model and five were added in recessive model severally. Moreover, the re-calculated OR value and corresponding 95% CI remained stable (homozygous model: OR = 1.09, 95% CI: 0.88 -1.35; recessive model: OR = 1.15, 95% CI: 0.96-1.37).
Between-study Heterogeneity Analysis
Based on the criterion of heterogeneity P-value greater than 0.1, significant heterogeneity was observed in all five genetic models (allele model Ph< 0.01; homozygous model Ph < 0.01; heterozygous model Ph = 0.06; dominant model Ph < 0.01; recessive model Ph < 0.01). Therefore, all of the analyses were conducted by applying the random effects model.
Since heterogeneity in the subgroup could not be completely eliminated in subgroup analyses, therefore, we drew Galbraith plots to find out the source of heterogeneity between studies. Eight studies [24-26,30,33,34,36,39] in the allele model, seven studies [24-27,32,36,39] in the homozygous model, one study  in the heterozygous model ( Figure 6 ), five studies in the dominant model, and six studies [24-26, 32,36,39] in the recessive model have been identified as the main causes of between-study heterogeneity.
After removing these data sets, no significant heterogeneity was detected (allele model Ph = 0.75; homozygous model Ph = 0.20; heterozygous model Ph = 0.31; dominant model Ph = 0.36; recessive model Ph = 0.46). The recomputed combined estimates also remained stable which meant the results were unchanged (allele model: OR= 1.01, 95% CI: 0.95 – 1.07; homozygous model: OR = 1.09, 95% CI: 0.97-1.22; heterozygous model: OR = 1.03, 95% CI: 0.96 – 1.11; dominant model: OR = 0.99, 95% CI: 0.92 – 1.08; recessive model: OR= 1.07，95％CI：0.97-1.17).