2.5. Statistical analyses

First, pairwise meta-analysis was conducted for a direct comparison of the different treatments using a RevMan 5.3 software (The Nordic Cochrane Center, The Cochrane Collaboration, Copenhagen, Denmark). A DerSimonian-Laird random effects model was employed to evaluate the standardized mean difference (SMD), odds ratio (OR) and its 95% confidence intervals (CI). χ² test and I² squared test were used to assess the heterogeneity. Subsequently, a Bayesian random effects model network meta-analysis was performed using a GeMTC 0.14.3 software (http://www.drugis.org/software/addis1/gemtc). We used the Markov Chains Monte Carlo (MCMC) method to calculate the results. For each outcome, the consistency model was applied that was based on 100,000 simulation iterations for each of the four chains. The tuning iterations were set as 50,000, and the thinning interval was 10. The Bayesian approach and the surface under the cumulative ranking (SUCRA) were used to calculate the probabilities of treatment ranking. Conversely, node-splitting analysis was used to estimate the inconsistency in the network meta-analysis. The plots of network and SUCRA were generated by using a STATA 15 (StataCorp, College Station, TX) software.