Stata software for windows version 12

Hardy-Weinberg Equilibrium (HWE) for each study was assessed using Chi-square test in control groups at a significance level of p<0.05. I² statistic and Cochrane Q test were used to measure heterogeneity. I² <50% or p>0.10 for the Q test indicated a lack of heterogeneity. The strength of association between COL1A2 rs42524 polymorphism and IA risk was calculated by odds ratio (OR) with 95% confidence interval (CI) using random effects model. The following genetic models were adopted: allelic contrast (C vs. G), homozygote contrast (CC vs. GG), heterozygote contrast (GC vs. GG), dominant model (CC + GC vs. GG), and recessive model (CC vs. GC +GG). Subgroup analyses were performed to explore the impact of the following: confounding factors: ethnicity, IA type and HWE. Sensitivity analysis was conducted by excluding each study one by one to challenge the robustness of pooling results. We used Begg’s and Egger’s tests to estimate possible publication bias [17 (link)]. All statistical analysis was performed using Stata software for Windows version 12.0 (Stata corporation, college station, TX, USA). P<0.05 was considered statistically significant.

The HWE of each study was tested using the chi-square test. Based on previous studies,16 (link) the ORs, and 95% confidence intervals (CIs) were calculated under the recessive genetic model (i.e., GG compared to CC+CG). The Z test and p<0.05 were used to examine the statistical significance of the pooled ORs.
Cochran’s Q-test and the I² test were used to evaluate the heterogeneity between the studies.17 (link),18 (link) If p≥0.1 of the Q-test, or I²≤50%, we considered that no significant heterogeneity existed, so the fixed effects model (Mantel-Haenszel method) was used in this meta-analysis. If not, the random effects model (DerSimonian and Laird method) would be considered. A subgroup analysis by ethnicity or genotype of HCV was performed to find sources of heterogeneity.
Sensitivity was determined by repeating the analysis and evaluating the stability of each result, after omitting in turn the data of each study. To assess the potential publication bias, we used Begger’s funnel plot and Egger’s linear regression test.19 (link) Stata software for Windows version 12.0 (Stata, College Station, TX, USA) was used for all the statistical analyses.

Continuous variables were expressed as median and interquartile (25th and 75th percentile) range. Categorical variables were expressed as frequencies and percentages. Categorical variables were compared with the Fisher’s exact probability test and the medians were compared using the Mann–Whitney U test. To compare more than two categorical variables, the chi-square (χ²) test was used. Statistical analyses were performed with STATA software for Windows, version 12.0 (Statacorp LP, College Station, TX, USA) with two-sided hypothesis testing and a P < 0.05 as the criteria for statistical significance.