Data extraction and quality assessment
Data extraction and presentation for the preparation of this manuscript followed the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses.Data abstraction was conducted by three authors (Qianwei Huang,Tian Zheng, Qianghui Huang), who independently used a predefined, standardized protocol, and data collection instrument. The following study-, patient-, and procedure-related data were extracted from the main paper. Any discrepancies were resolved by discussion among the authors. The Newcastle Ottawa Quality Assessment Scale (NOS) for observational trials was used to evaluate the quality of the included studies.
The odds ratio (OR) with 95% confidence intervals (CIs) was used to analyze dichotomous data, whereas weighted mean differences were used to analyze continuous variables. The Cochran Q test was used to assess the presence of heterogeneity among studies, and the Higgins I2 test was used to assess the extent of the observed heterogeneity, which suggests 0-25%, 25-50%, and 50-75% as low, moderate, and high heterogeneity, respectively. Given the inherent differences in study design and demographics, we used a random-effects model to estimate the pooled effects of our meta-analyses.
For the primary endpoints, prespecified sensitivity analyses were performed by iteratively removing one study at a time to confirm that no single study had a significant impact on our results. To further test the stability of our meta analysis, we performed multiple subgroup analyses (based on study group,different surgical methods). A random-effects meta-regression analysis was performed to examine the effect of preselected covariates on the overall effect. In both groups, the logarithm of OR for the primary endpoints was regressed against age, percentage of hypertension, diabetes mellitus, stroke, left ventricular ejection fraction (LVEF), and size of left atrial (LA) diameter.
We examined the asymmetry of the funnel plots in detail and further assessed them using the Begg adjusted rank correlation test and the Egger regression asymmetry test to detect any publication bias in the primary endpoints. All statistical analyses were carried out using the RevMan software package (Review Manager, Version 5.3) and STATA software 12.0. P-values were two-tailed, with 0.05 considered statistically significant.