Two researchers independently screened the titles and abstracts to evaluate the potential relevance of studies. Disagreements were resolved by consensus or discussion with a third author. After screening , full-text reviews were performed. Detailed study information, interventions, controls, and outcomes were retrieved using a standardized data extraction protocol.Statistical analysis was conducted with Review Manager software .
Chi-square test was applied to test heterogeneity , and if it was found to exist, sensitivity analysis was applied to find out the cause. A random-effects model was used again if heterogeneity still could not be eliminated. Sensitivity analysis, subgroup analysis, or just descriptive analysis were applied when meeting significant clinical heterogeneity. A fixed-effects model was used for data with no significant heterogeneity .